<?xml version="1.0" encoding="UTF-8"?><rss xmlns:dc="http://purl.org/dc/elements/1.1/" xmlns:content="http://purl.org/rss/1.0/modules/content/" xmlns:atom="http://www.w3.org/2005/Atom" version="2.0" xmlns:cc="http://cyber.law.harvard.edu/rss/creativeCommonsRssModule.html">
    <channel>
        <title><![CDATA[Stories by Vikas Pujar on Medium]]></title>
        <description><![CDATA[Stories by Vikas Pujar on Medium]]></description>
        <link>https://medium.com/@vikaspo?source=rss-db0a0da6d449------2</link>
        <image>
            <url>https://cdn-images-1.medium.com/fit/c/150/150/0*y9UF9XNHnk6Ec1hm</url>
            <title>Stories by Vikas Pujar on Medium</title>
            <link>https://medium.com/@vikaspo?source=rss-db0a0da6d449------2</link>
        </image>
        <generator>Medium</generator>
        <lastBuildDate>Tue, 17 Nov 2026 17:07:27 GMT</lastBuildDate>
        <atom:link href="https://medium.com/@vikaspo/feed" rel="self" type="application/rss+xml"/>
        <webMaster><![CDATA[yourfriends@medium.com]]></webMaster>
        <atom:link href="http://medium.superfeedr.com/" rel="hub"/>
        <item>
            <title><![CDATA[Cache me if you can! Rethinking Mainframe Data Caching]]></title>
            <link>https://medium.com/elevate-tech/cache-me-if-you-can-rethinking-mainframe-data-caching-73964777a526?source=rss-db0a0da6d449------2</link>
            <guid isPermaLink="false">https://medium.com/p/73964777a526</guid>
            <category><![CDATA[mainframe]]></category>
            <category><![CDATA[system-of-record]]></category>
            <category><![CDATA[caching]]></category>
            <category><![CDATA[ibm-zdih]]></category>
            <category><![CDATA[data-modernization]]></category>
            <dc:creator><![CDATA[Vikas Pujar]]></dc:creator>
            <pubDate>Tue, 10 Nov 2026 16:27:14 GMT</pubDate>
            <atom:updated>2026-11-11T10:08:30.769Z</atom:updated>
            <content:encoded><![CDATA[<figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*FhYFfbBp-AWQU22Lm3HbZQ.png" /><figcaption>Teaser image for “Cache me if you can! Rethinking mainframe data caching”. Read-only enquiries such as balance, order status and policy details go to a cache that answers from memory. The cache sits inside a mainframe (IBM Z) next to the system of record (Db2, IMS, VSAM), and only a cache miss goes to the system of record.</figcaption></figure><h3>Chapter 1 : Introduction</h3><p>Think of a typical digital application. A customer opens the mobile app to check account details. The app calls the mainframe, CICS or IMS reads Db2, IMS or VSAM, does some processing and sends back the response.</p><p>Now think of salary day. I check my bank account again and again to see if the salary is credited or not :-). I am sure many others do the same. Thousands of customers are checking their balance many times within a few minutes.</p><p>This raises an important question: do we really need to go to the System of Record for every request? Most of these requests only read data, they do not change it.</p><p>One common option is to offload the data outside the mainframe, for example to a distributed database, and serve these read-only requests from there. But what if the mainframe shop does not want the data to move out of the mainframe? What is the alternate solution?</p><p>This is where caching comes in. Caching is not a new idea. Redis, Hazelcast, browser cache and CDN all work on the same principle: keep frequently used data closer to the consumer. For mainframe data, the cache can be kept close to the System of Record, so the data can stay within the mainframe environment. But this brings a bigger question: how much data should we copy into a cache, and what are we ready to give up in return? There is also a catch. If the cache is not updated on time, I will not see my salary, and I will call the bank.</p><p>In my earlier blogs, I covered data decoupling, Digital Integration Hub and data virtualization (DVM). Caching is closely related to these topics. In this blog, I will look at when caching makes sense and when it does not.</p><h3>Chapter 2 : Why Cache Mainframe Data?</h3><p>As discussed in the previous chapter, there are also other ways to offload read-only data. But why should mainframe shops look at caching? Below are some of the reasons (not a complete list).</p><ul><li><strong>Less load on the System of Record:</strong> When fewer requests reach Db2 or VSAM, they have more capacity to serve other requests, so overall performance improves. Take customer profile data like name, address, status and segment. It is read thousands of times but changes only occasionally. Going to Db2 every time is not needed.</li><li><strong>MIPS optimization:</strong> Every data access call uses CPU in CICS, Db2 and SQL execution. If the cache answers the request, this work is avoided, and the CPU (MIPS) is freed up for the workloads that need it.</li><li><strong>Lower latency:</strong> The request does not need to travel through z/OS, CICS or IMS and Db2 every time. An in-memory lookup is faster.</li><li><strong>Better scalability:</strong> Many applications can read from the cache instead of all of them hitting the System of Record.</li><li><strong>Data stays within the IBM Z environment:</strong> Compared with offloading to a distributed database, the copy can stay on the mainframe platform, so existing security and compliance controls can still apply. Please note that this depends on where the cache runs.</li></ul><h3>Chapter 3 : The Catch, to Cache or not to Cache</h3><p>Here is an honest question: “Should I cache this data or not?” I have to admit that I don’t have a ready answer. Yes, it depends on many variables, but it still takes me time to understand them properly. This is one of the reasons I wrote this blog.</p><p>Let’s go back to salary day. My balance check is exactly the kind of data I would put in the high-risk group below. If Db2 shows the salary as credited, but the cache still shows the old balance, I will call the bank. So the real question is: how stale can the data be?</p><p>Here is how I look at different data, sample ones and non -exhaustive.</p><ul><li><strong>Good fit:</strong> Product catalogue (minutes to hours), branch information (hours), customer profile (seconds to minutes).</li><li><strong>Handle with care:</strong> Account summary (seconds), available credit (near real-time).</li><li><strong>High risk, avoid:</strong> Account balance, payment transaction, fraud decision.</li></ul><p>So the question is not “Should we cache?” The question is: what data, for how long, and under which consistency model? By consistency model, I mean the rule for how fresh the cache must be. It can be strong (always the latest data) or eventual (can lag for a short time). The flowchart below helps with the first step, deciding whether a piece of data is a good candidate for caching.</p><figure><img alt="Decision flowchart for caching mainframe data. Five yes/no questions in order: Is the data read frequently? Is the data fairly stable? Is some staleness acceptable? Is System of Record access costly enough? Can we keep the cache in sync? If all answers are Yes, caching may be a good fit. If any answer is No, optimize the System of Record access first." src="https://cdn-images-1.medium.com/max/1024/1*C_I-DSVfIZ-qzPnPjLPhqQ.png" /><figcaption>If all five answers are Yes, caching may be a good fit. If any answer is No, optimize the System of Record access first.</figcaption></figure><p>Now, let’s come back to the salary-day example. One idea is to update the cache from an event the moment the credit is posted. This keeps the cache almost as fresh as the System of Record, but only if the synchronization is very reliable. Also, if the cache has to be updated every second, then there is little point in caching that data, because the cache does extra work for very few reads between two updates. So deciding which data to cache and which not to cache is a very important first step.</p><h3>Chapter 4 : Keeping the Cache in Sync, and the Real Cost</h3><p>Once you decide what to cache, there are four common ways to keep it in sync.</p><ul><li><strong>Cache-aside:</strong> The application checks the cache first. If data is missing, it reads the System of Record and fills the cache. Simple, but data can go stale.</li><li><strong>Write-through:</strong> The update goes to the cache and the System of Record together. Better consistency, but tighter coupling.</li><li><strong>Write-behind:</strong> The cache accepts the write and updates the System of Record later. Fast, but now we must think about durability and failures.</li><li><strong>Event-driven:</strong> CICS or IMS publish transaction events, and the cache is updated from those events. This suits IBM Z well, and I cover it in Chapter 5.</li></ul><p>Caching helps in MIPS optimization only when it removes costly and frequent work from the main path. Let’s take an example. There are 1 million requests, and the cache hit ratio is 95%. Only about 50,000 requests will reach Db2 for that access pattern. Please note that the 95% hit ratio is only an example. Actual savings depend on your workload and access pattern. A cache is not free performance.</p><p>These are the hidden costs architects often forget.</p><ul><li><strong>Memory:</strong> Cached data can be much bigger than the original record.</li><li><strong>Cost of change capture:</strong> Capturing and publishing changes also uses CPU on the mainframe.</li><li><strong>High availability:</strong> If a cache node fails, do you rebuild, reload, fall back to Db2, or keep replicas?</li><li><strong>Cache warming:</strong> After a restart, the cache is empty. Millions of requests then go straight to Db2. Your performance fix becomes a Db2 storm.</li><li><strong>Cache eviction:</strong> When memory is full, what is removed first? Oldest first, time-based expiry, business priority or size?</li><li><strong>Synchronization:</strong> How fast does a change reach the cache? This is often the hardest part.</li><li><strong>Security:</strong> A cache holds a copy of sensitive data, so the same access control, encryption and audit rules must apply. If security is the reason you avoid offload, the cache must meet the same standard.</li></ul><h3>Chapter 5 : Caching Solutions on the Mainframe</h3><p>Here are three solutions worth knowing.</p><ul><li><strong>IBM zDIH (IBM Z Digital Integration Hub):</strong> I covered it in my earlier blog (link : <a href="https://medium.com/@vikaspo/exploring-the-role-of-ibm-zdih-in-mainframe-hybrid-cloud-integration-52c3d2a75842">https://medium.com/@vikaspo/exploring-the-role-of-ibm-zdih-in-mainframe-hybrid-cloud-integration-52c3d2a75842</a>). IBM describes it as a way to share real-time information from core Systems of Record on z/OS with hybrid cloud applications [1]. It follows the event-driven pattern. CICS, IMS and batch programs share their changes as events or log streams, zDIH builds the in-memory cache from these, and applications read from the cache through SQL, REST and connectors like Kafka.</li><li><strong>QuickSelect for Db2 (EasiRun):</strong> It keeps the results of repeated, rarely changing SELECT and FETCH statements in memory, so Db2 does not run the same query again and again. It works for Db2 data only. Its survey mode can help estimate the savings before you decide.</li><li><strong>Hazelcast:</strong> It adds a distributed in-memory data layer beside IBM Z, kept current by change data capture, so inquiry reads are served from memory while CICS and COBOL run unchanged. It can serve many applications, and it can also bring in non-mainframe sources. It can run on Linux on Z or LinuxONE.</li></ul><p>In simple words, QuickSelect caches at the Db2 level, while zDIH and Hazelcast build a separate in-memory data layer that can serve many applications.</p><h3>Chapter 6 : Final Thoughts</h3><p>Caching is an architecture decision, not a shortcut. First decide whether this data needs to be cached at all, and if yes, look at the benefits and then at the architecture concerns like freshness, memory and synchronization. Only after that, look at the mainframe solutions like zDIH, QuickSelect, Hazelcast or other solutions like this.</p><p>Unlike my other blogs, where I shared a roadmap or solution and mostly closed the topic, I am keeping this one open. I feel this topic still needs inputs from mainframe practitioners like you, and I am eager to hear your views. What data do you cache on the mainframe, and what do you avoid?</p><p><strong>References</strong></p><p>[1] <a href="https://www.ibm.com/support/z-content-solutions/z-digital-integration-hub/">https://www.ibm.com/support/z-content-solutions/z-digital-integration-hub/</a></p><p>[2] <a href="https://easirun.com/products/quickselect-for-db2/">https://easirun.com/products/quickselect-for-db2/</a></p><p>[3]<a href="https://hazelcast.com/blog/ibm-linuxone-hybrid-cloud-ai-savings/">https://hazelcast.com/blog/ibm-linuxone-hybrid-cloud-ai-savings/</a></p><p><em>Disclaimer:</em></p><ol><li><em>The views and opinions expressed in this blog are solely those of the author and do not necessarily reflect the official policy or position of the author’s employer or any other organization. The information provided is based on personal experience and research and is intended for informational purposes only.</em></li><li><em>Blog is written by me, but enhanced by LLM :)</em></li></ol><img src="https://medium.com/_/stat?event=post.clientViewed&referrerSource=full_rss&postId=73964777a526" width="1" height="1" alt=""><hr><p><a href="https://medium.com/elevate-tech/cache-me-if-you-can-rethinking-mainframe-data-caching-73964777a526">Cache me if you can! Rethinking Mainframe Data Caching</a> was originally published in <a href="https://medium.com/elevate-tech">Elevate Tech</a> on Medium, where people are continuing the conversation by highlighting and responding to this story.</p>]]></content:encoded>
        </item>
        <item>
            <title><![CDATA[IBM Bob Premium Package for Z — A Practical View for Mainframe Teams]]></title>
            <link>https://medium.com/elevate-tech/ibm-bob-premium-package-for-z-a-practical-view-for-mainframe-teams-c33cea973aaf?source=rss-db0a0da6d449------2</link>
            <guid isPermaLink="false">https://medium.com/p/c33cea973aaf</guid>
            <category><![CDATA[mainframe]]></category>
            <category><![CDATA[ibm-bob]]></category>
            <category><![CDATA[mainframe-modernization]]></category>
            <category><![CDATA[agentic-ai]]></category>
            <category><![CDATA[cobol]]></category>
            <dc:creator><![CDATA[Vikas Pujar]]></dc:creator>
            <pubDate>Mon, 27 Jul 2026 15:09:49 GMT</pubDate>
            <atom:updated>2026-08-04T05:31:45.476Z</atom:updated>
            <content:encoded><![CDATA[<blockquote>A practical, hands-on review of IBM’s Agentic AI assistant for modern mainframe development.</blockquote><figure><img alt="Agentic AI Robot standing and helping in creating of Cobol code and providing metrics around Cobol Code. The technologies of mainframe like Cobol, JCL, DB2, VSAN, CICS and modern language Java and Agentic AI also listed." src="https://cdn-images-1.medium.com/max/1024/1*_OeKBBKr6zNVveqjwklQeA.jpeg" /><figcaption>Bob for Z — Agentic IDE for Mainframe SDLC ( Image Credit — ChatGPT)</figcaption></figure><h3>Chapter 1: Introduction</h3><p>The mainframe world has changed a lot in the last few years. We have moved from green-screen development to Eclipse-based tools like IDz, and now to agentic IDEs like IBM Bob. Each step has improved productivity, but the move to AI-assisted development feels like paradigm shift. I wrote in my earlier blog about how Agentic AI is changing the IT landscape — <a href="https://medium.com/elevate-tech/from-automation-to-autonomy-how-agentic-ai-can-transform-the-mainframe-landscape-ce6a4ed023d4">Agentic AI Blog.</a> Now will extend it to Mainframe focussed Agentic AI tool from IBM — <strong>IBM Bob Premium Package for z</strong>.</p><p>IBM recently introduced the <strong>IBM Bob Premium Package for z</strong>, which builds on watsonx Code Assistant for z and brings deeper IBM z awareness into the Bob experience. In this blog, I will share what this package is, what it can do, and some practical views from my own experience using it.</p><h3>Chapter 2: What Is IBM Bob Premium Package for Z?</h3><p>Before exploring the Premium Package for z, lets talk about IBM Bob. Introduced at IBM TechXchange 2025 and made generally available in March 2026, IBM Bob is an AI-powered software development assistant which under the hood uses best of the breed models (LLMs) that help teams understand, generate, review, refactor, and modernize code across the software development lifecycle. While IBM Bob primarily focuses on distributed application development, the Premium Package for z extends these capabilities to IBM Z and mainframe environments.</p><p>At its core, the <strong>IBM Bob Premium Package for z</strong> is an AI-powered extension for the IBM Bob IDE and Bob Shell, designed specifically for mainframe applications. It helps teams analyze, document, refactor, and modernize COBOL, PL/I , REXX, HLASM etc codebases. Unlike generic AI tools, it understands IBM z context — including middleware like CICS, IMS, DB2, and MQ — and uses that knowledge to provide more relevant explanations, suggestions, and code changes ( I extend on how its different in next para).</p><p>For me, the biggest USP of Bob for z is the <strong>Understand</strong> feature. Unlike standard agentic IDEs, which usually take code as input and then start analysis from scratch using prompts or markdown files, Bob for z already begins with a better understanding of the application code. This is possible because of ADDI static analysis, which builds metadata about how modules interact with each other, how many copybooks a program depends on, and similar details. Because of this, Understand can give a much better view of the application. Adding link to my old blog where I talk in detail about mainframe discovery tool ADDI ( Which is now zUnderstand) — <a href="http://medium.com/@vikaspo/mainframe-application-discovery-4974d04f1e20">http://medium.com/@vikaspo/mainframe-application-discovery-4974d04f1e20</a>.</p><p>Mainframe developers / users spend a significant amount of time understanding legacy applications before making even small changes. Business logic is often undocumented, dependencies span hundreds of programs, and experienced SMEs are becoming increasingly scarce. AI-assisted development tools such as <strong>IBM Bob Premium Package for z</strong> aim to reduce this effort by helping developers understand, analyze, and modify legacy applications more efficiently.</p><p>Bob for z was generally made available from June 24th 2026. But I had opportunity to explore Bob for z premium as early access user (perk being IBM champion) from April mid and will share my hands on experience in next chapters.</p><h3>Chapter 3: Key Capabilities of Bob for Z</h3><p>The package is built around two main modes: z <strong>Architect Mode</strong> and z <strong>Code Mode</strong>.</p><h4>1. z Architect Mode</h4><p>This mode is for system-level reasoning. It helps you:</p><ul><li>Understand application structure and dependencies across large COBOL and PL/I applications</li><li>Identify business intent and change impact before making updates</li><li>Plan modernization with a clearer view of risks and priorities</li></ul><p>For architects and technical leads, this is where the package starts to feel different from simple code assistants.</p><h4>2. z Code Mode</h4><p>This mode focuses on developer tasks. It can:</p><ul><li>Generate and refactor COBOL, PL/I, JCL, and other mainframe code aligned to coding standards</li><li>Optimize and debug code using Z-aware context</li><li>Reduce technical debt and time to delivery through prioritized insights</li></ul><p>For developers, it is like having an AI co-developer that understands mainframe conventions and dependencies.</p><h3>Chapter 4: Workflows and Data Dictionary</h3><p>One of the strong points of the Premium Package is its <strong>workflow-driven approach</strong>. Instead of random chat-based suggestions, you get guided workflows for common modernization tasks:</p><ul><li>Generate documentation for COBOL, PL/I, and Assembler programs</li><li>Extract business rules from legacy code</li><li>Build and manage a <strong>data dictionary</strong> for variables across the application</li><li>Refactor COBOL or PL/I programs into modular, reusable services</li></ul><p>The <strong>data dictionary</strong> is particularly useful. It gives a semantic view of variables ( which are often not so self explanatory), including names, descriptions, and detailed explanations. As you approve or adjust AI-generated entries, the dictionary grows and becomes more accurate for your specific application.</p><h3>Chapter 5: My Experience So Far with Bob for z Premium Package</h3><p>As an early user of IBM Bob Premium Package for Z, I got the chance to explore its features and see how it can help mainframe teams in their day-to-day work. Please note that Bob itself is an IDE (VsCode fork) and comes with all features of a standard eclipse IDE. You need Bob for extension VSIX files to explore z premium features.</p><p>The first step is to load the source code to be analyzed into a local folder or clone it from a Git repository. In my case, I loaded the COBOL code into a local folder and opened it in the BOB IDE.</p><p>Next, I executed the /init command in the BOB chat window. This triggered a series of initialization tasks, including scanning the COBOL codebase. Based on its analysis, BOB automatically generated an AGENTS.md file, which provides the AI agent with the necessary context about the application.</p><p>In addition, BOB’s zUnderstand capability scans the codebase and builds metadata, which is stored in a database such as Microsoft SQL Server. This metadata can then be accessed through an MCP server, enabling BOB to perform deeper code understanding and dependency analysis. Although this step is not shown in the screenshot below, it forms an important part of the overall architecture.</p><p>Post this , I used <strong>Bob</strong> to implement a standard change request on a COBOL batch program. The program reads customer records from a bank file, filters specific bank accounts, sorts them, and writes the results to an output file. The enhancement request was straightforward — the bank code used for filtering needed to be expanded from <strong>4 bytes to 6 bytes</strong>.</p><p>Before making the change, I used <strong>Architect Mode</strong> to understand the existing application logic. Within seconds, Bob provided a detailed explanation of the program’s functionality, including its processing flow, business logic, and key data transformations. The screenshot below shows the comprehensive analysis generated by Bob, helping me confidently identify where the required change needed to be implemented.</p><figure><img alt="Bob for Eclipse-based IDEs lets users select COBOL code and ask the agent to explain what it does. The agent analyzes the code with backend LLMs and returns a natural-language explanation in English or another preferred language. In image a Cobol code is selected and Bob has provided explanation of code in english." src="https://cdn-images-1.medium.com/max/1024/1*msiZLg6V0V02VvCVDmGFFQ.png" /><figcaption>Bob for z — Providing code explanation</figcaption></figure><p>Next, I provided Bob with the following change request:</p><p><em>“I have a change request to modify the variable </em><strong><em>BANK-CODE</em></strong><em> from </em><strong><em>4 bytes to 6 bytes</em></strong><em>. Please analyze the application repository and identify all COBOL programs, copybooks, JCLs, and other dependent artifacts that need to be updated to implement this change.”</em></p><p>Bob analyzed the repository, traced the usage of the <strong>BANK-CODE</strong> field across the application, and identified the impacted programs and copybooks. This significantly reduced the manual effort typically required for impact analysis and provided a clear starting point for implementing the change.</p><p>As shown in the image below, Bob creates a list of tasks it needs to perform for the change and then executes them one after the other. It also asks for human approval after each step, so you can review the changes and provide feedback if something is not right.</p><figure><img alt="In Bob Eclipse-based IDEs, a user asks an agent to make a COBOL code change, and the agent breaks the request into subtasks, plans the work, and then carries it out with approval before modifying the source code. In image for the ask to expand the Cobol variable BNK-CODE from 4 bytes to 6 bytes and Bob lists all the tasks it has to carry out to perform this change. To first task to search copybooks where BNK-CODE variable is used to last task to verification of all changes." src="https://cdn-images-1.medium.com/max/1024/1*ZLjEKHzYuel14r6k1CyNIA.png" /><figcaption>Bob for z — Listing steps to perform the code change</figcaption></figure><p>Once the code change is done, Bob provides a comparison report of the new code versus the old code.</p><figure><img alt="Bob eclipse based IDE, once code change is done , Bob compare old code vs new code and lists the changes made. In image Bob compares the old code where BANK-CODE File length is increased by 2 bytes and variable length increased by 2 bytes , where as old code still has original length." src="https://cdn-images-1.medium.com/max/1024/1*pJJDinZoYhIwAZKqh4xQrw.png" /><figcaption>Bob for z — comparing the old vs new code after code change, helps in review of code</figcaption></figure><p>At the end, it provides test cases and scenarios to check the changes performed. I am hopeful that in the future we will have IBM TAZ integration with the Bob IDE, so these tests can also be executed by connecting directly to z/OS.</p><figure><img alt="Bob for z eclipse based IDE where agent lists unit test cases needed to test the changes performed. In image the Agentic AI IDe Bob adds task of creating test cases to its existing list and and then performs the task and lists unit test cases to test the code change of increasing BANK-CODE variable from 4 bytes to 6 bytes." src="https://cdn-images-1.medium.com/max/1024/1*bdjW8U7HqcwB6xyGiQ3dzQ.png" /><figcaption>Bob for z — Creating test cases</figcaption></figure><p>But keep in mind , since it relatively new offering, note that :</p><ul><li>Some prompts required refinement.</li><li>A few dependency recommendations needed manual validation.</li><li>Human review is still essential before implementing production changes.</li></ul><h3>Chapter 6: Where It Fits in Modernization</h3><p>If you are thinking about In-Place mainframe modernization, this tool fits well in several scenarios, few of them listed below (non-exhaustive):</p><ul><li>You want to <strong>understand</strong> a large, complex COBOL codebase before making changes</li><li>You need to <strong>document</strong> business rules and logic for audit, compliance, or knowledge transfer</li><li>You plan to <strong>refactor</strong> or modularize applications but want to do it in a controlled, traceable way</li><li>You want an AI co-developer that is <strong>mainframe-aware</strong>, not just generic code AI</li></ul><p>and list goes on…</p><h3>Chapter 7: Final Thoughts</h3><p>As mentioned in my opening remarks, IBM Bob Premium Package for z represents a significant shift in the way mainframe applications are developed and maintained. It has the potential to make developers far more productive and efficient. For someone like me, who started their career coding on green screens with very limited automation, having an Agentic AI assistant for mainframe development feels like moving from a horse carriage to a Ferrari! 🙂</p><p><strong>References</strong></p><ul><li>Introducing the IBM Bob Premium Package for Z<br>https://www.ibm.com/new/announcements/announcing-the-private-early-access-preview-of-the-ibm-bob-premium-package-for-z</li><li>IBM Bob Premium Package for Z documentation<br>https://www.ibm.com/docs/en/bobz/3.0.0</li><li>IBM Bob Premium Package for Z — Overview and workflows<br>https://bob.ibm.com/docs/ide/premium-packages/bob-for-z/bob-for-z-index</li></ul><img src="https://medium.com/_/stat?event=post.clientViewed&referrerSource=full_rss&postId=c33cea973aaf" width="1" height="1" alt=""><hr><p><a href="https://medium.com/elevate-tech/ibm-bob-premium-package-for-z-a-practical-view-for-mainframe-teams-c33cea973aaf">IBM Bob Premium Package for Z — A Practical View for Mainframe Teams</a> was originally published in <a href="https://medium.com/elevate-tech">Elevate Tech</a> on Medium, where people are continuing the conversation by highlighting and responding to this story.</p>]]></content:encoded>
        </item>
        <item>
            <title><![CDATA[COBOL and Containers: A Complicated Relationship]]></title>
            <link>https://medium.com/elevate-tech/cobol-and-containers-a-complicated-relationship-1fae395b44d8?source=rss-db0a0da6d449------2</link>
            <guid isPermaLink="false">https://medium.com/p/1fae395b44d8</guid>
            <category><![CDATA[containers]]></category>
            <category><![CDATA[cobol]]></category>
            <category><![CDATA[legacy-modernization]]></category>
            <category><![CDATA[ibm-z]]></category>
            <category><![CDATA[mainframe]]></category>
            <dc:creator><![CDATA[Vikas Pujar]]></dc:creator>
            <pubDate>Mon, 15 Jun 2026 15:56:09 GMT</pubDate>
            <atom:updated>2026-07-30T05:25:59.395Z</atom:updated>
            <content:encoded><![CDATA[<p>Exploring whether containerization is the right modernization strategy for Enterprise COBOL applications running on IBM Z and z/OS.</p><figure><img alt="Image showing one side as Mainframe, running technologies like COBOL, DB2, VSAM, IMS, and JCL/JES, and representing decades of reliability for mission-critical workloads. On the other side is Container/Kubernetes, running API gateways, web applications, monitoring services, and AI/ML workloads, showing why clients use it for cloud-native, modern, agile, and scalable applications. Finally, aspects to consider like Business value, Performance, Risk and Complexity and is it the right solution?" src="https://cdn-images-1.medium.com/max/1024/1*B0skjdoAqg7YY0NaF4m4mg.png" /><figcaption>Mainframe and its core technologies, benefits vs Containerization and its core technologies , benefits. Image Source — GPT</figcaption></figure><h3>Chapter 1: Introduction</h3><p>Containers are not new. Docker and Kubernetes have been around for more than ten years. In many ways, they have changed how we build, deploy, and manage modern applications. They have become a standard part of cloud-native development and help teams deliver software faster and more consistently across different environments.</p><p>At the same time, COBOL — first released more than 65 years ago — still powers some of the most critical business systems in the world. Banks, insurance companies, and governments running on IBM Z use it to process millions of transactions and batch jobs every day.</p><p>A common question I often hear (and sometimes ask myself) is: <strong>Can COBOL applications be containerized?</strong> In this blog, I will explore this topic.</p><h3>Chapter 2: COBOL Overview</h3><p>COBOL stands for Common Business-Oriented Language. It was designed in 1959 by a group called CODASYL. The language was created specifically for business data processing — tasks like reading records, performing calculations, and generating reports. Its syntax reads almost like plain English on purpose. The goal was that even non-programmers, such as business managers, could understand the code.</p><p>Over the years, several COBOL variants have appeared. Some popular ones include IBM VS COBOL, COBOL II, Enterprise COBOL for z/OS, Micro Focus COBOL, and GnuCOBOL. Each version has its own features, runtime, and platform requirements.</p><p>In this blog, I will mainly focus on <strong>IBM Enterprise COBOL on z/OS</strong>, because it runs the majority of enterprise COBOL workloads in production today. At the end, I will also briefly discuss GnuCOBOL, which has a very different story when it comes to containers.</p><h3>Chapter 3: Containers Overview</h3><p>Containers became popular because of technologies like Docker and Kubernetes. The idea is simple. A container packages an application together with its runtime, libraries, and other dependencies, so it runs in the same way across different environments.</p><p>Containers offer several potential benefits:</p><ul><li>Faster deployment</li><li>Consistent behavior across environments</li><li>Easy scaling</li><li>Simpler DevOps process</li><li>Better use of system resources</li></ul><p>It is important to note that container technology became popular as a way to simplify the deployment, portability, and scalability of distributed applications such as Java, Python, and Node.js services. These characteristics are very different from traditional COBOL applications, which are often tightly coupled to mainframe services and data.</p><h3>Chapter 4: Can Enterprise COBOL Applications Be Containerized?</h3><p>The answer is not a simple yes or no. It depends on the COBOL variant and the platform.</p><p>But here is the more important question: <strong>Not “Can you?” but “Should you?”</strong></p><p>With the rise of containerization and the shift away from standalone VM-based deployments, containerization was often seen as a form of modernization. It also created the impression that moving an application into a container is always the right next step. For distributed applications (if not for all of them), that may hold true. But for traditional COBOL workloads on z/OS, the situation is not that simple.</p><p>Traditional Enterprise COBOL applications running natively on z/OS cannot be containerized in the same way as modern Java or Python applications because they depend heavily on z/OS subsystems such as CICS, IMS, JES, VSAM, and DB2 for z/OS. These services cannot be easily packaged into a container without major changes to the application. The short answer to why you cannot do it today is that the z/OS platform currently does not support containerization of COBOL applications running on z/OS address space, due to the limitations mentioned above. (Note : zOS allows containerization of other workloads which I cover in zCX chapter 6 below).</p><p>That said, not all COBOL is the same. COBOL running on Linux (using GnuCOBOL or other distributed runtimes) <strong>can</strong> be containerized. I will cover that in the next chapter.</p><p>Even if it is technically possible, is it a good idea for COBOL z/OS workloads? Here is my view on why it is usually <strong>not</strong> the best fit, I would be interested to hear your thoughts in the comments.</p><ul><li><strong>The “it works on my machine” problem doesn’t exist the same way.</strong> One of the original motivations behind containers was to eliminate environment inconsistencies between development, test, and production. On IBM Z, workloads typically run on a highly standardized and centrally managed platform, reducing many of the deployment inconsistencies that containers were designed to solve.</li><li><strong>COBOL applications are rarely standalone.</strong> A single transaction often depends on many copybooks, subprograms, JCL procedures, control cards, databases, and other services. Containers work best with small, self-contained, microservices architecture apps. Packaging a monolithic application that has hundreds of dependencies, along with a shared data layer used by multiple applications, is not simple. In many cases, the challenge is not just about packaging the code, but also about handling all the internal dependencies and shared data interactions that the application relies on which makes it complex — almost like running a virtual machine again.</li><li><strong>Performance and I/O matter a lot.</strong> Many COBOL batch jobs process millions of records with heavy I/O to DB2, VSAM, or IMS. On the mainframe, the application and data are tightly optimized and close together. Moving data-intensive batch workloads away from the tightly integrated z/OS environment may introduce additional network, storage, or architectural layers that should be carefully evaluated against existing SLA requirements. For example , consider a nightly interest calculation batch job that processes 50 million customer records from DB2 and VSAM. The application is optimized for sequential I/O and runs within a tightly integrated z/OS environment. Simply placing the COBOL code inside a container does not automatically improve scalability, agility, or operational efficiency.</li><li><strong>The mainframe already has many container-like benefits.</strong> Many of the capabilities that organizations seek from container platforms — such as workload isolation, resource management, scalability, security, and high availability — have existed on IBM Z for decades through technologies such as LPARs, Workload Manager (WLM), Parallel Sysplex, RACF, and JES.</li></ul><h3>Chapter 5: GnuCOBOL and Containerization</h3><p>The story is quite different for GnuCOBOL. GnuCOBOL applications run on Linux and can be packaged inside Docker containers just like any other application. For example:</p><ul><li>GnuCOBOL runtime</li><li>COBOL application</li><li>Configuration files</li></ul><p>…can all be bundled into a container image and deployed using Kubernetes.</p><p>In these environments, containerization can provide real benefits — portability, cloud deployment flexibility, and easier DevOps integration.</p><p>However, it is important to recognize that GnuCOBOL running on Linux is fundamentally different from Enterprise COBOL running on IBM Z. GnuCOBOL was released in 2002, at a time when service-oriented architecture was already becoming common. Many applications were therefore designed with clearer separation of functions, and since they ran on x86 servers that could scale horizontally, their architecture was already more aligned with the platform they ran on. This is one reason why GnuCOBOL is generally more suitable for containerization.</p><p>I have also shared a hands-on demonstration of GnuCOBOL running in a container on LinkedIn. Feel free to take a look link — <a href="https://www.linkedin.com/posts/vikaspujar_gnucobol-also-known-as-opencobol-is-an-activity-7164219565986566144--YBd">have a look here</a>.</p><h3>Chapter 6: What About z/OS Container Extensions (zCX)?</h3><p>If you have been following discussions around containers on IBM Z, you have probably heard about zCX . IBM introduced zCX to bring container technology to the IBM Z platform. But it is important to understand what it is actually designed for.</p><p>zCX is built to run Linux-based container workloads <em>within</em> a z/OS environment. It is not meant for running traditional Enterprise COBOL applications inside containers. That is a common misunderstanding.</p><p>In reality, zCX is mainly intended for modern workloads such as API gateways, web applications, open-source tools, AI services, and monitoring platforms. The main objective is to bring cloud-native applications <em>closer</em> to enterprise data on IBM Z — a co-location model, not a COBOL containerization model. Traditional Enterprise COBOL applications continue to run natively on z/OS in LPARs, not inside containers. In short : zCX brings containers to IBM Z; it does not bring Enterprise COBOL into containers.</p><h3>Conclusion</h3><p>The question is not whether COBOL can be containerized. The question is whether containerization solves a real business problem. Modernization should be driven by business value, architecture, and workload characteristics — not by technology trends alone.</p><p>References :</p><ol><li>800 Lines of COBOL Source : <a href="https://www.rocketsoftware.com/sites/default/files/resource_files/modernize-on-strength.pdf">https://www.rocketsoftware.com/sites/default/files/resource_files/modernize-on-strength.pdf</a></li></ol><p><em>Disclaimer:</em></p><ol><li><em>The views and opinions expressed in this blog are solely those of the author and do not necessarily reflect the official policy or position of the author’s employer or any other organization. The information provided is based on personal experience and research and is intended for informational purposes only.</em></li><li><em>Blog is written by me with my words and thoughts, but enhanced by LLM :)</em></li></ol><img src="https://medium.com/_/stat?event=post.clientViewed&referrerSource=full_rss&postId=1fae395b44d8" width="1" height="1" alt=""><hr><p><a href="https://medium.com/elevate-tech/cobol-and-containers-a-complicated-relationship-1fae395b44d8">COBOL and Containers: A Complicated Relationship</a> was originally published in <a href="https://medium.com/elevate-tech">Elevate Tech</a> on Medium, where people are continuing the conversation by highlighting and responding to this story.</p>]]></content:encoded>
        </item>
        <item>
            <title><![CDATA[Data Virtualization: Simplifying Mainframe Data Access Across Sequential Files, VSAM, and DB2]]></title>
            <link>https://medium.com/elevate-tech/data-virtualization-simplifying-mainframe-data-access-across-sequential-files-vsam-and-db2-1a26c4c1da0d?source=rss-db0a0da6d449------2</link>
            <guid isPermaLink="false">https://medium.com/p/1a26c4c1da0d</guid>
            <dc:creator><![CDATA[Vikas Pujar]]></dc:creator>
            <pubDate>Mon, 25 May 2026 20:20:25 GMT</pubDate>
            <atom:updated>2026-06-01T13:59:58.300Z</atom:updated>
            <content:encoded><![CDATA[<figure><img alt="The mainframe servers in data center connected to each other" src="https://cdn-images-1.medium.com/max/1024/1*Yi6gtmkvWx6rbNNdaH6AxQ.avif" /><figcaption>Mainframe Server in data center</figcaption></figure><p><strong>Chapter 1 : Introduction</strong></p><p>Mainframe data is often seen as “dark data” because it is not easily accessible to users outside the mainframe environment. Even today, many mission-critical COBOL applications store data different data stored on mainframe like Sequential files, GDGs, VSAM datasets, and DB2,IMS tables. According to a 2024 survey, nearly 70% of enterprise data still resides on mainframe systems. [1]</p><p>This raises an important question: how can we expose or access this so-called dark data?</p><p>In my earlier blogs, I discussed some of the available approaches to make mainframe data accessible to external systems:</p><ul><li><a href="https://medium.com/elevate-tech/breaking-data-silos-modernization-frameworks-for-ibm-z-mainframe-data-stores-4b52ab4187a9">Data Decoupling Solutions</a></li><li><a href="https://vikaspo.medium.com/exploring-the-role-of-ibm-zdih-in-mainframe-hybrid-cloud-integration-52c3d2a75842">Digital Integration Hub</a></li></ul><p>However, the most common suggestion you will hear is: why not replicate the data from the mainframe to external data stores like PostgreSQL or other distributed databases?</p><p>But what if replication is not an option? Many organizations avoid it due to security, compliance, latency, or cost concerns. In such cases, the requirement is clear: keep the data on the mainframe, but still make it accessible to external applications.</p><p>This is where IBM Data Virtualization Manager (DVM) becomes highly relevant.</p><p>In this blog, I will share my experience with IBM DVM for z/OS, explain how it works, and show how it enables access to different types of mainframe data through a single virtualization layer.</p><p>Please note that there are other solutions available as well, such as Rocket’s data virtualization offering for mainframe data stores. However, in this blog, I will focus mainly on IBM’s DVM.</p><p><strong>Chapter 2 : What is IBM Data Virtualization Manager for zOS?</strong></p><p>IBM Data Virtualization Manager (DVM) is an IBM solution that provides real-time virtual access to different types of data sources on the mainframe. DVM works like a virtual layer between mainframe data and the outside world. It takes data from sources like VSAM, sequential files, DB2, and other supported stores, and presents them in a simple SQL form, so applications can read them without moving or copying the data.</p><p>The image below shows different types of data stores available on the mainframe, though this is not an exhaustive list, and how DVM helps expose them to the external world.</p><figure><img alt="DVM takes data from multiple mainframe data stores such as DB2 and IDMS etc on mainframe and views are creted out of these data stores. These views are managed on the virtualization server and then exposed to external applications like ML and AI through JDBC and ODBC connections. Tha data stays on mainframe which is important aspect of this offering." src="https://cdn-images-1.medium.com/max/1024/1*fiJdISaqFJRm8Hm65CCoUg.png" /><figcaption>IBM Data Virtulization Manager Architecture — Source: Image created by GPT with my inputs</figcaption></figure><p>DVM maps the mainframe data into virtual tables and views. After that, users, APIs, reporting tools, and modern applications can query the data in real time, almost like it is coming from a relational database.</p><p>This means:</p><ul><li>No need for complex ETL processes</li><li>No duplicate copies of data</li><li>Faster access to enterprise data</li><li>Reduced data movement</li><li>Better integration with modern applications</li></ul><p>One thing I personally like about DVM is that it provides access to almost every type of data store on the mainframe, as shown in the image above. Both application teams and operations teams can leverage it for their respective use cases.</p><p><strong>Chapter 3 : How to use IBM Data Virtualization Manager</strong></p><p>DVM acts as a bridge between modern applications and traditional mainframe data sources.</p><p>Let’s take a banking example.</p><p>In many banks, customer data is spread across multiple systems. For example, account balance and loan details may be stored in DB2 tables, while customer profile information like address and contact details may be stored in VSAM datasets.</p><p>Now assume there is a business requirement. The bank wants to identify customers who are eligible for a pre-approved loan offer based on their current balance and also needs their contact details to send notifications.</p><p>In this case:</p><ul><li>DB2 table contains: Customer ID, Account Balance, Loan Eligibility Score</li><li>VSAM dataset contains: Customer ID, Name, Mobile Number, Address</li></ul><p>Using DVM, we can solve this without moving any data.</p><p>First, create a virtual table on the DB2 table to fetch required columns like Customer ID and Account Balance (or eligibility score).</p><p>Next, create a virtual table on the VSAM dataset to fetch Customer ID, Name, and Mobile Number.</p><p>Below image shows DVM where you create virtual table and select source for same , in this example for first virtual table I select DBMS (DB2) and second virtual table I select VSAM.</p><figure><img alt="The Eclipse based IDE for DVM where selecting a DB2 as data source and creating virtual table from it. The process is simple — just right-click on the DB2 table and select the required columns, then those cols are used to create virtual table. This Virtual table is used down the line to create virtual view." src="https://cdn-images-1.medium.com/max/1024/1*eBxKCZOG-P89Cyr8v3busQ.png" /><figcaption>IBM DVM Tool — Create Table View. Source: DVM Trial Access Screenshot</figcaption></figure><figure><img alt="A wizard window of DVM that allows to select the data type by either typing the name of selecting from a list. The list contains the entries ADABAS, DBMS , IDMS, IMS, Logstream, MQ, Sequential Files, VSAM, VSAM CICS, zFS. Users select data type of their choice and proceed with creation of virtual table." src="https://cdn-images-1.medium.com/max/698/1*E712J6zuQ8tlYP6e1JYRCw.png" /><figcaption>IBM DVM Tool — Select Data Source View. Source: DVM Trial Access Screenshot</figcaption></figure><p>Finally, create a combined virtual view by joining these two views on Customer ID.</p><figure><img alt="The Eclipse based IDE for DVM where virtual tables created in earlier steps are used to create virtual view . For example, a virtual table from DB2 and a virtual table from VSAM created in previous steps can be combined into one virtual view. Thie view is then exposed to outer world thru JDBC , ODBC connections." src="https://cdn-images-1.medium.com/max/1024/1*_OInjr4vaJcYpzzkd6LBCg.png" /><figcaption>IBM DVM Tool — Create Virtual View View. Source: DVM Trail Access Screenshot</figcaption></figure><p>Now the business team can directly query this final virtual view to:</p><ul><li>Identify eligible customers</li><li>Get their contact details</li><li>Trigger campaigns or notifications</li></ul><p>All this happens in real time, without copying data or building batch jobs.</p><p>With this example, you should get a basic idea of how DVM can be used.</p><p>In the above examples, I have used screenshots from an older version of DVM, but the overall concept and message remain the same.</p><p><strong>Chapter 4 : What are the benefits of using DVM</strong></p><p>The benefits of DVM can be grouped into two areas: technical benefits and business benefits.</p><p><strong>Technical Benefits</strong></p><ul><li>No data movement required<br> Data stays on the mainframe, no need for replication</li><li>Real-time data access<br> Latest data is available instantly, no dependency on batch jobs</li><li>Reduced ETL effort<br> No need to build and maintain complex ETL pipelines</li><li>Single unified view<br> Access data from VSAM, DB2, and sequential files in one place</li><li>Faster development<br> Developers can use standard SQL instead of custom programs</li><li>Easy integration<br> Works well with APIs, analytics tools, and cloud applications</li></ul><p><strong>Business Benefits</strong></p><ul><li>Better data security<br> Data remains on mainframe with existing security and compliance controls</li><li>Lower operational cost<br> No extra infrastructure needed for data duplication and pipelines</li><li>Faster decision making<br> Real-time access helps business teams act quickly</li><li>Supports modernization<br> Enables new digital use cases without major system changes</li><li>Improved data accessibility<br> Makes mainframe data easily available to different teams</li></ul><p>Please note that any underlying changes to mainframe data stores, such as DB2 schema changes, are not automatically picked up by DVM. In such cases, remapping of fields is needed to ensure the virtual views and tables continue to work correctly.</p><p><strong>Chapter 5 : Use of DVM in other products</strong></p><p>DB2 SQL Insights allows you to use DB2 tables for building machine learning models. As the name suggests, it works well when your data is already in DB2.</p><p>But what if you need to use other mainframe data sources as well? This is where DVM fits in. You can use DVM to create views on data stores like VSAM, GDG, and other mainframe data sources, and then use that data for machine learning models.</p><p><strong>Chapter 6 : Final Thoughts</strong></p><p>Mainframes still continue to be the system of record for many enterprises. The real challenge today is not about lack of data, but how quickly and efficiently we can access and use that data.</p><p>DVM provides a practical approach by enabling real-time access to mutiple data sources on mainframe without physically moving the data. From my experience working with large mainframe shops, I see DVM as a very practical solution to bridge traditional mainframe systems with modern digital architectures. It helps organizations modernize data access while still leveraging the stability and reliability of the mainframe.</p><p>P.S. I have kept this blog simple and provided a high-level view of DVM. If you are interested in more technical details, I strongly recommend going through the IBM Redbook on DVM: <a href="https://www.redbooks.ibm.com/redbooks/pdfs/sg248514.pdf">IBM Redbook on DVM</a></p><p>[1]<a href="https://www.forbes.com/sites/heatherwishartsmith/2024/11/12/mainframes-the-backbone-of-the-worldwide-economy/">https://www.forbes.com/sites/heatherwishartsmith/2024/11/12/mainframes-the-backbone-of-the-worldwide-economy/</a></p><p><em>Disclaimer:</em></p><ol><li><em>The views and opinions expressed in this blog are solely those of the author and do not necessarily reflect the official policy or position of the author’s employer or any other organization. The information provided is based on personal experience and research and is intended for informational purposes only.</em></li><li><em>Blog is written by me , but enhanced by LLM :)</em></li></ol><img src="https://medium.com/_/stat?event=post.clientViewed&referrerSource=full_rss&postId=1a26c4c1da0d" width="1" height="1" alt=""><hr><p><a href="https://medium.com/elevate-tech/data-virtualization-simplifying-mainframe-data-access-across-sequential-files-vsam-and-db2-1a26c4c1da0d">Data Virtualization: Simplifying Mainframe Data Access Across Sequential Files, VSAM, and DB2</a> was originally published in <a href="https://medium.com/elevate-tech">Elevate Tech</a> on Medium, where people are continuing the conversation by highlighting and responding to this story.</p>]]></content:encoded>
        </item>
        <item>
            <title><![CDATA[Understanding Mainframe Processors: CP, IFL, zIIP, and AI Chips Demystified]]></title>
            <link>https://medium.com/elevate-tech/understanding-mainframe-processors-cp-ifl-ziip-and-ai-chips-demystified-bde493dac304?source=rss-db0a0da6d449------2</link>
            <guid isPermaLink="false">https://medium.com/p/bde493dac304</guid>
            <category><![CDATA[mainframe]]></category>
            <category><![CDATA[mainframe-modernization]]></category>
            <category><![CDATA[ai]]></category>
            <category><![CDATA[processors]]></category>
            <category><![CDATA[genai]]></category>
            <dc:creator><![CDATA[Vikas Pujar]]></dc:creator>
            <pubDate>Tue, 06 Jan 2026 13:57:52 GMT</pubDate>
            <atom:updated>2026-01-07T15:44:46.947Z</atom:updated>
            <content:encoded><![CDATA[<h4>.</h4><h4><strong>Chapter 1 : Introduction</strong></h4><p>Let me first wish all my readers a very happy New Year 2026! Over the past year, I have written multiple blogs on the latest mainframe offerings, especially around AI on Z, including Generative AI, and how these capabilities are shaping the future of the platform. I have also shared my views on the new AI chips, such as Telum and Spyre in my earlier blogs.</p><p>One topic that often creates confusion in the mainframe world is the different processor types. Terms like “General CP,” “IFL,” and “ZIIP” are sometimes used interchangeably, which leads to misunderstandings. In this post, I will focus on the different types of processors available on the mainframe and explain how they work together to support both traditional and modern workloads.</p><h4>Chapter 2: Overview of CPUs in a Mainframe</h4><p>Before getting into the topic, if you are new to the mainframe world, I would recommend going through my earlier blog that introduces the mainframe platform and explains the different operating systems it can run (link : <a href="https://medium.com/@vikaspo/is-mainframe-available-on-cloud-5bdae6c7d489">https://medium.com/@vikaspo/is-mainframe-available-on-cloud-5bdae6c7d489</a>). This will help you build foundation and make it easier to understand the concepts discussed further in this blog.</p><p>IBM Z systems processors/chips can be broadly classified into four categories:</p><ol><li>General Purpose Central Processors<strong> (</strong>CPs)</li><li>Speciality Engines like z Integrated Information Processor (zIIP) and z<strong> </strong>Application Assist Processor (zAAP)</li><li>Integrated Facility for Linux (IFL)</li><li>AI accelerators (both on-chip and attached)</li></ol><figure><img alt="" src="https://cdn-images-1.medium.com/max/635/1*AezXGVkU0xxuFQOvCJ1Ibg.png" /><figcaption>Depicting different LPARs with different Operating Systems and different processors, Image Source : ijfmr.com</figcaption></figure><p>Each of these processor types is designed to handle specific types of workloads. It is important to note that all of these processors coexist within a single IBM Z system as shown in image above.</p><h4>Chapter 3: What is General CP and Its Purpose</h4><p>The General Central Processor (CP) is the primary processor used by z/OS to run traditional mainframe workloads. Any workload executed on this processor is considered for software billing. (Software- or MIPS-based billing is a separate topic in itself, will take it up some other time).</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/406/1*2WXLR2Uee8rsPU-BMVaY7A.png" /><figcaption>Processor Unit Chip , Image source : IBM</figcaption></figure><p>Typical workloads that can be run on CPs include (non-exhaustive)</p><ul><li>COBOL, PL/I, Assembler workloads</li><li>CICS, IMS, DB2 core processing</li><li>JES, system tasks, batch jobs, etc.</li></ul><p>It is possible for an IBM Z system to be configured with only general-purpose CPs. In such a setup, all workloads are executed on CPs, with no offloading to other processors (discussed in the next points).</p><h4>Chapter 4: What is zIIP and Its Purpose</h4><p>zIIPs are specialty processors designed to offload eligible workloads from general-purpose CPs. These include modern workloads such as Java and Python, along with selected database and middleware processing, helping optimize software costs while maintaining required performance. As mentioned earlier, zIIPs are dependent on CPs for fallback execution when workloads are not eligible or when zIIP capacity is unavailable. Please note that newer systems (z13 onwards) support “<a href="https://www.google.com/search?q=zAAP+on+zIIP&amp;rlz=1C1GCEA_enDE1179DE1179&amp;oq=zIIP+and+zAAP+same%3F&amp;gs_lcrp=EgZjaHJvbWUyBggAEEUYOdIBCDQwOTVqMGo3qAIAsAIA&amp;sourceid=chrome&amp;ie=UTF-8&amp;mstk=AUtExfDIuKntgpVoMdCsxRT-z7k4VL5Y0voryujcYc7vdwvXOQ9DuwWk9s8gkfWheBNd83rfJlFCFzjvAT3J9BkG2_B-QxX3eQqF62k5Lyqli5xrOW-SvgLiRJ083te61eg1f-yNqTnMZGfvfgKoQZLh268EqT7yO82aNdyFtZ6UzX-Q9rAiPcYzadSyxKwbGz6N32TLJe29z244dw0mNpdMKbvGligVfkdgNytJOAyB8MnaR6yrhtms4LvgUtdVMAQk4IY_yAgGfE3bM-jau90_L7bnktsiBRGbD4esldvT4LCSlA&amp;csui=3&amp;ved=2ahUKEwiBlo6E5PaRAxXDQ_EDHc5YOkcQgK4QegQIARAC">zAAP on zIIP</a>,” allowing zIIPs to run zAAP-eligible work, making them functionally unified for many users. Hence not covering zAAP seperatly in this blog.</p><p>Typical workloads that can be run on zIIPs include (non-exhaustive)</p><ul><li>Java workloads on z/OS</li><li>Select DB2 SQL processing</li><li>Analytics, AI inferences</li><li>Select Python on z/OS workloads, etc.</li></ul><p>One obvious question is whether we can add any number of zIIP engines and offload majority of workloads to them. The answer is no. There are specific eligibility rules and ratio constraints between CPs and zIIPs that govern how many zIIP engines can be configured on a system. It is also important to note that zIIPs do not replace CPs — they complement them by efficiently offloading eligible and modern workloads while still relying on CPs for core processing.</p><h4>Chapter 5: What is IFL and Its Purpose</h4><p>Integrated Facility for Linux (IFL) is a dedicated processor designed specifically to run Linux workloads on IBM Z and LinuxONE systems. Note that it’s complemented by virtualization technologies such as IBM z/VM and KVM on Z, enabling efficient and scalable Linux deployments on the mainframe. IFLs also help in cost optimization since workloads executed on IFLs do not directly contribute to software billing.</p><p>Typical workloads that can be run on IFLs include (non-exhaustive)</p><ul><li>Linux on Z</li><li>Red Hat OpenShift</li><li>Java, Python, Node.js on Linux, etc.</li></ul><p>One common question I often get is whether container workloads running on Linux on Z — such as OpenShift — can be simply lift-and-shifted to x86 platforms for scenarios like cloud bursting. The short answer is no. The underlying CPU architectures are different. If you visit Docker Hub and check the supported architectures for container images, you’ll notice that many images are built specifically for the IBM Z (s390x) architecture. These images cannot run directly on x86 systems, which makes a straight lift-and-shift approach not feasible.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/280/1*vyobVyBN9m6LlEdaBjvf5Q.png" /><figcaption>Architecture Types on DockerHub Image Source — DockerHub</figcaption></figure><h4>Chapter 6: What are AI Accelerators and Its Purpose on IBM Z</h4><p>IBM introduced dedicated AI processors with the z16 hardware in the form of the Telum chip. Telum I was launched with z16, and with z17 IBM further extended AI capabilities by introducing the Spyre accelerator and Telum II chips. Telum processors are designed for real-time, in-transaction AI inferencing, while Spyre focuses on enabling Generative AI workloads on IBM Z.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*ojrAD98MDVEdw2h0u4GHdA.png" /><figcaption>Telum Chip , Image source IBM</figcaption></figure><p>Typical use cases that can leverage AI chips include (non-exhaustive)</p><ul><li>Fraud detection</li><li>Risk scoring</li><li>GenAI LLM workloads, etc.</li></ul><p>As per IBM, the z17 can process up to 450 billion inference operations per day using multiple AI models for use cases like credit card fraud detection.</p><p>With multi-model AI becoming the norm, these chips help run both core AI and Generative AI workloads directly on the mainframe, close to the data and critical business transactions.</p><h4>Chapter 7: Conclusion</h4><p>The “right workload for the right platform” message can be extended to “the right processor for the right workload.” With the range of processor options discussed in the blog, mainframe teams can strategically leverage each to maximize value and performance. As adoption of Java, Python, APIs, OpenShift, and AI continues to grow on IBM Z, utilization of zIIP, Telum, Spyre, and IFL processors will rise accordingly — further redefining the mainframe’s role in the enterprise hybrid cloud era.</p><p>References :</p><p>[1] <a href="https://www.ibm.com/new/announcements/ibm-spyre-accelerator-and-telum-ii-processor-capturing-ai-value-at-a-trusted-enterprise-level">https://www.ibm.com/new/announcements/ibm-spyre-accelerator-and-telum-ii-processor-capturing-ai-value-at-a-trusted-enterprise-level</a></p><p>[2] <a href="https://www.ijfmr.com/papers/2020/3/22604.pdf">https://www.ijfmr.com/papers/2020/3/22604.pdf</a></p><p><em>Disclaimer:</em></p><ol><li><em>The views and opinions expressed in this blog are solely those of the author and do not necessarily reflect the official policy or position of the author’s employer or any other organization. The information provided is based on personal experience and research and is intended for informational purposes only.</em></li><li><em>Blog is written by me , but enhanced by LLM :)</em></li></ol><img src="https://medium.com/_/stat?event=post.clientViewed&referrerSource=full_rss&postId=bde493dac304" width="1" height="1" alt=""><hr><p><a href="https://medium.com/elevate-tech/understanding-mainframe-processors-cp-ifl-ziip-and-ai-chips-demystified-bde493dac304">Understanding Mainframe Processors: CP, IFL, zIIP, and AI Chips Demystified</a> was originally published in <a href="https://medium.com/elevate-tech">Elevate Tech</a> on Medium, where people are continuing the conversation by highlighting and responding to this story.</p>]]></content:encoded>
        </item>
        <item>
            <title><![CDATA[From Automation to Autonomy: How Agentic AI Can Transform the Mainframe Landscape]]></title>
            <link>https://medium.com/elevate-tech/from-automation-to-autonomy-how-agentic-ai-can-transform-the-mainframe-landscape-ce6a4ed023d4?source=rss-db0a0da6d449------2</link>
            <guid isPermaLink="false">https://medium.com/p/ce6a4ed023d4</guid>
            <category><![CDATA[agentic-ai]]></category>
            <category><![CDATA[mainframe-modernization]]></category>
            <category><![CDATA[mainframe]]></category>
            <category><![CDATA[elevate-tech]]></category>
            <category><![CDATA[genai]]></category>
            <dc:creator><![CDATA[Vikas Pujar]]></dc:creator>
            <pubDate>Mon, 03 Nov 2025 11:15:44 GMT</pubDate>
            <atom:updated>2026-01-08T06:21:57.800Z</atom:updated>
            <content:encoded><![CDATA[<figure><img alt="" src="https://cdn-images-1.medium.com/proxy/1*dkeiX5dAW2p4DlGnMmmnJQ.png" /><figcaption>Mainframe offerings are now being integrated with Agentic AI Framework ( Image source : GPT Created Image)</figcaption></figure><p><strong>Chapter 1 : Introduction</strong></p><p>Mainframe computing and artificial intelligence share deep historical roots. The term Artificial Intelligence was first coined in 1956 by John McCarthy at the Dartmouth Conference, marking the formal beginning of AI research. Just eight years later, in 1964, IBM introduced its first mainframe system, revolutionizing enterprise computing. These two foundational technologies — born in the same era — are now converging powerfully on platforms like the IBM z16 and z17, enabling secure, scalable, and AI-infused enterprise operations.</p><p>Large Language Models (LLMs) and Generative AI have captured significant attention since their emergence. Now, the focus is shifting to the next major evolution in AI — Agentic AI — which emphasizes autonomy, decision-making, and goal-driven behavior. This blog explores the concept of Agentic AI and how it can be effectively adopted within the mainframe ecosystem.</p><p><strong>Chapter 2 : What is Agentic AI</strong></p><p>At its core, Agentic AI is a type of AI that’s all about autonomy. This means that it can make decisions, take actions, and even learn on its own to achieve specific goals. It’s kind of like having a virtual assistant that can think, reason, and adapt to changing circumstances without needing constant direction.</p><p><strong>Agentic AI operates in four key stages</strong></p><p>a) Perception: It gathers data from the world around it.</p><p>b) Reasoning: It processes this data to understand what’s going on.</p><p>c) Action: It decides what to do based on its understanding.</p><p>d) Learning: It improves and adapts over time, learning from feedback and experience.</p><p>In discussions about Agentic AI, you’ll often encounter the term ‘AI agents.’ While closely related, these terms have distinct meanings. Agentic AI is the overall concept of artificial intelligence systems that can act independently and achieve goals. AI-powered agents are the individual components within the system that execute tasks.</p><p><strong>Agentic AI vs Gen AI vs Traditional AI</strong></p><p>AI encompasses multiple subfields, and it’s important to understand the differences between them</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/553/1*vDR0Oqg_cw1APy2oQ5OrMw.png" /><figcaption>Comparision of Agentic AI, GenAI and Traditional AI</figcaption></figure><p><strong>Chapter 3 : Agentic AI and Mainframe</strong></p><p>At the time of this writing, many mainframe vendors are actively exploring Agentic AI frameworks. While there are multiple potential use cases, the initial focus is largely on operations, where the availability of large volumes of raw data makes it a natural fit for workflow automation through Agentic AI. For example listed below two offerings (Operations and non-exhaustive) in market today :</p><p><strong>IBM :</strong> IBM has introduced agentic AI capabilities in the latest version of IBM watsonx Assistant for Z, specifically designed to simplify IT operations for IBM Z.</p><p>At TechXchange 2025, IBM unveiled <strong>Project Bob</strong>, tool for agentic AI-driven enterprise software development. Designed to act as an intelligent development partner, Project Bob interprets developer intent, navigates entire code repositories, and enforces compliance standards — ensuring high-quality code with every commit. While Project Bob has not yet been announced for IBM Z, hope we get similar framework for IBM Z in near future.</p><p><strong>BMC :</strong> BMC has announced <em>HelixGPT</em> (<a href="https://www.bmc.com/it-solutions/helixgpt.html?utm_source=chatgpt.com">https://www.bmc.com/it-solutions/helixgpt.html</a>), a solution that unifies Service Management and IT Operations (AIOps) into a single platform for mainframes powered by Agentic AI.</p><p>New offerings are expected to expand rapidly in the near future, signaling the beginning of a broader shift toward autonomous and intelligent mainframe development and operations leveraging Agentic AI.</p><p><strong>Chapter 4 : Potential cases on the mainframe include (non-exhaustive)</strong></p><p>Mainframe use cases for Agentic AI span across operations, development, compliance etc. ranging from autonomous monitoring and workload automation to code analysis, natural language explanations, and even automated code transformation.</p><p>a) AIOps: Autonomous monitoring of logs, SMF data, and job activity to detect anomalies, recommend or apply fixes, and prevent failures.</p><p>b) Workload automation: AI agents dynamically adjust job schedules, resolving dependencies, and managing capacity based on real-time inputs.</p><p>c) DevOps acceleration: Agents scanning for code changes, triggering CI/CD pipelines, running quality checks, and auto-generating documentation.</p><p>d) Code Analysis and Understanding: AI agents can scan COBOL, JCL, assembler code and identify logical modules, unused code, and dependencies. Natural language agents can explain business logic in plain English, making it easier for modern developers to understand the intent behind legacy programs.</p><p><strong>Conclusion </strong>:</p><p>Nearly 70 years after IBM introduced its first mainframe, the platform continues to power mission-critical workloads across banking, insurance, and government. Now, as AI shifts from automation to autonomy, mainframes are once again at the center of technological reinvention — this time, led by Agentic AI.</p><p>Mainframes have always embraced emerging technologies, and today we see Large Language Models (LLMs) playing an increasingly important role in the mainframe ecosystem. As Agentic AI matures, it promises to transform mainframes from powerful systems into intelligent, autonomous platforms that enhance efficiency, accelerate innovation, and deliver long-term business value.</p><p>References :</p><p>[1] <a href="https://www.ibm.com/new/announcements/empowering-mainframe-users-with-agentic-ai">https://www.ibm.com/new/announcements/empowering-mainframe-users-with-agentic-ai</a></p><p>[2] <a href="https://www.iteuropa.com/news/bmc-helix-aims-grow-partner-base-around-ai-agentic-services">https://www.iteuropa.com/news/bmc-helix-aims-grow-partner-base-around-ai-agentic-services</a></p><p><em>Disclaimer:</em></p><ol><li><em>The views and opinions expressed in this blog are solely those of the author and do not necessarily reflect the official policy or position of the author’s employer or any other organization. The information provided is based on personal experience and research and is intended for informational purposes only.</em></li><li><em>Blog is written by me , but enhanced by LLM :)</em></li></ol><img src="https://medium.com/_/stat?event=post.clientViewed&referrerSource=full_rss&postId=ce6a4ed023d4" width="1" height="1" alt=""><hr><p><a href="https://medium.com/elevate-tech/from-automation-to-autonomy-how-agentic-ai-can-transform-the-mainframe-landscape-ce6a4ed023d4">From Automation to Autonomy: How Agentic AI Can Transform the Mainframe Landscape</a> was originally published in <a href="https://medium.com/elevate-tech">Elevate Tech</a> on Medium, where people are continuing the conversation by highlighting and responding to this story.</p>]]></content:encoded>
        </item>
        <item>
            <title><![CDATA[Eccox Tool APT]]></title>
            <link>https://vikaspo.medium.com/eccox-tool-apt-4604e6956892?source=rss-db0a0da6d449------2</link>
            <guid isPermaLink="false">https://medium.com/p/4604e6956892</guid>
            <dc:creator><![CDATA[Vikas Pujar]]></dc:creator>
            <pubDate>Thu, 25 Sep 2025 13:22:47 GMT</pubDate>
            <atom:updated>2026-01-19T20:53:15.463Z</atom:updated>
            <content:encoded><![CDATA[<p><strong>Introduction</strong></p><p>Testing is a critical component of the Software Development Life Cycle (SDLC), typically accounting for 20–30% of the overall effort and going up to 50% in case of mainframe modernization testing. However, in mainframe systems suitable test environments are not readily available, also building and maintaining new test environments is often costly and complex. A commonly used approach has been to replicate the entire environment across multiple logical partitions (LPARs), enabling multiple test sessions to run in parallel, each with its own set of resources.</p><p>However, this approach requires additional hardware resources to be allocated to each LPAR, along with considerable effort from system programmers to set up and maintain and manage these LPARs.</p><p>Imagine an application being tested in isolation within an LPAR with its own code and data. This is precisely where the Eccox APT tool comes in. In this POV, we’ll explore the capabilities and benefits of the APT tool.</p><p><strong>Eccox Application for Parallel Testing (Eccox APT) Overview</strong></p><p>mEccox APT is a test environment management solution for mainframe clients. APT sets up, manages, and maintains isolated test environments quickly and efficiently. APT’s primary innovation lies in creating Logical APT z/OS Containers — virtualized test environments with code and data necessary for testing specific application changes. <em>It is important to emphasize that this isolation takes place </em><strong><em>within the LPAR</em></strong>, where the Dev and QA teams perform testing of applications.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/442/1*fMxubK9uvTdTgvNwTS1Y1w.jpeg" /></figure><p>As shown in the image above, the APT Clones real components necessary for functional tests, like load modules, DB2 tables, files, and JCL, are isolated in environments on the same Mainframe LPAR, and all of this is done with WebUI. Components that are not changed during processing can be shared with other applications and do not need to be replicated.</p><p><strong>Need for Eccox APT solution on Mainframe</strong></p><p>The maximum number of mainframes LPARS is limited to 85 per zMachine, which limits the number of possible test environments considering production-to-test LPAR ratio up to 90–10%. Apart from that spinning of new TEST LPARs is complex ,costly and time consuming and hence typically not pursued by clients on demand.</p><p>One of the biggest challenges of testing applications in existing development and testing LPARs is shared code and data, typically from parallel projects. Clients often look for an isolated test environment that facilitates testing of project-specific changes without interference from undesirable code versions, modifications from other projects, or data corruption.</p><p>Also, the quality of code to be deployed to production is paramount to the client, and shift-left testing in isolated environment allows them to identify the issues early in the SDLC. Eccox addresses all these challenges.</p><p><strong>Key Features of Eccox APT</strong>:</p><p>1. Logical APT z/OS Isolated test environments : This setup supports parallel test execution and multiple testing sessions within a single LPAR, reducing the need for additional hardware and enhancing resource utilization.</p><p>2. Web Interface for Simplified Test Configuration : With Web UI, teams can easily configure test cases, provision test environments, and manage the entire testing lifecycle.</p><p>3. Automated Test Environment Management : No manual configurations or delays in recreating, updating, or reprocessing tests and dependencies. Eccox APT automates the provision of isolated runtime test environments, reducing setup time by up to 90%.</p><p>4. Application discovery feature : APT discovers the code and its dependencies required to create an isolated environment. For example, for online applications by providing transaction name APT can identify all its dependencies (programs, DBs, files, etc.).</p><p>5. File and Database Management : APT provides a feature to copy a file and database from different environments (testbeds or LPARs) into the APT dedicated environment.</p><p>6. Golden copy / Snapshots : The APT provides the ability to save the current state of the APT-dedicated environment, including structure and data; later, users can restore databases to a previously known state even after modifications, deletions, or alterations.</p><p>7. Integration with DevOps : APT can retrieve code from popular Git frameworks (GitLab, Bitbucket, Azure DevOps, etc.) and integrate with schedulers like Control-M to search and schedule jobs from the scheduler.</p><p><strong>Benefits of Eccox APT tool (non-exhaustive)</strong></p><p>1. Accelerated Testing : Teams can create, modify, and rerun test environments quickly, enabling faster release cycles and supporting Agile, DevOps practices.</p><p>2. Increased Business Agility : Comprehensive testing in isolated environments significantly reduces the risk of defects in production, ensuring more reliable and secure software releases, resulting in improved time to market.</p><p>3. Cost Efficiency : By reducing the need for multiple LPARs and physical hardware, Eccox APT lowers the overall testing infrastructure costs.</p><p>4. Effective Resource Utilization : Eccox APT Containers maximize the use of existing LPARs, allowing for parallel testing without extra hardware investment.</p><p><em>5. </em>Test Coverage : Eccox APT helps in enhanced test coverage leading to improved quality of changes to be deployed into production.</p><p>6. Performance and Scalability: Eccox/APT leverages existing mainframe infrastructure to deliver performance and scalability, enabling organizations to support real runtime environments.</p><p><strong>Conclusion:</strong></p><p>Eccox APT allows testing in isolated and prod-like environments, ensuring that changes are validated across all dependencies between applications and data processing flows prior to deployment, reducing the risk of production failures and post-release fixes. As mainframe applications become increasingly integrated into agile and DevOps workflows, tools like Eccox APT are essential for maintaining the speed and quality of software delivery.</p><p>The Eccox APT solution is unique, and no other solution provides the offering to run dedicated test environments <strong>within</strong> an LPAR.</p><p><em>Reference: </em><a href="https://www.eccoxapt.com/"><em>https://www.eccoxapt.com/</em></a></p><p><em>Disclaimer:</em></p><ol><li><em>The views and opinions expressed in this blog are solely those of the author and do not necessarily reflect the official policy or position of the author’s employer or any other organization. The information provided is based on personal experience and research and is intended for informational purposes only.</em></li><li><em>Blog is written by me , but enhanced by LLM :)</em></li></ol><img src="https://medium.com/_/stat?event=post.clientViewed&referrerSource=full_rss&postId=4604e6956892" width="1" height="1" alt="">]]></content:encoded>
        </item>
        <item>
            <title><![CDATA[Future-Proofing Mainframe Development with Endevor-Git Integration]]></title>
            <link>https://vikaspo.medium.com/future-proofing-mainframe-development-with-endevor-git-integration-5f447fd0e217?source=rss-db0a0da6d449------2</link>
            <guid isPermaLink="false">https://medium.com/p/5f447fd0e217</guid>
            <dc:creator><![CDATA[Vikas Pujar]]></dc:creator>
            <pubDate>Mon, 22 Sep 2025 18:37:11 GMT</pubDate>
            <atom:updated>2025-09-22T18:37:11.752Z</atom:updated>
            <content:encoded><![CDATA[<p><strong>Introduction</strong></p><p>It’s no exaggeration to say the world runs on code. According to a Micro Focus survey [1], nearly 800 billion lines of COBOL are still in production today. That raises an obvious question: how is such a massive codebase maintained and managed? The answer lies in source code management (SCM) tools, also known as version control systems. SCM plays a critical role in tracking and managing code changes, ensuring teams always work on the right version. In this blog, we’ll explore how SCMs like Endevor and Git are used in the mainframe world.</p><p><strong>Chapter 1 : Mainframe and Distributed Systems SCMs</strong></p><p>Mainframe and distributed systems have traditionally relied on dedicated SCM tools, but in recent years we’ve seen a convergence driven by client demand to harmonize toolchains across platforms.</p><p>SCMs generally fall into two categories: centralized and decentralized. In a centralized model, all code resides in a single repository or server, while in a decentralized model, code can exist across multiple repositories or even on a developer’s machine.</p><p>At their core, SCM tools perform three major functions: code storage (versioning), code build (compilation), and code deployment. In distributed systems, these tasks are typically handled by separate tools — for example, Git for versioning, Gradle for builds, and Ansible for deployments. On the mainframe, however, single tools like Endevor and Changeman cover all three functions.</p><p>With the rise of DevOps, SCM tools have regained prominence, as organizations seek solutions that integrate seamlessly into their pipelines. This trend has further accelerated Git’s popularity as the SCM of choice in modern environments</p><p><strong>Chapter 2 :Mainframe and SCM Tools</strong><br>The mainframe technology stack includes a variety of components — such as COBOL, PL/I, JCL, and Assembler — and each of these must be versioned and managed through SCMs. In the mainframe market, the two most widely used SCM tools are <strong>Endevor</strong> and <strong>Changeman</strong>.</p><p>Endevor, a Broadcom (formerly CA) product, is adopted by large number of mainframe organizations. Owing to its centralized architecture, the entire software codebase is stored and maintained on a single server — specifically, the mainframe itself.</p><p><strong>Chapter 3 : Need for Distributed Source Code Management Tools</strong></p><p><strong>Challenge:</strong> Traditional SCMs like Endevor are centralized, which limits parallel development and makes integration with modern DevOps tools and pipelines difficult. This creates a roadblock for organizations aiming to fully embrace DevOps practices.</p><p><strong>Solution:</strong> The Broadcom Git Bridge addresses this challenge by enabling Git-based versioning while continuing to leverage Endevor for core functions such as build and deployment. This hybrid approach allows teams to work with familiar DevOps tools while maintaining the robustness of mainframe SCM.</p><p><strong>Chapter 4 : Benefits of Git as Versioning tool :</strong></p><p>1. Parallel Development: Git enables multiple developers to work simultaneously on a single source of code.</p><p>2. No code merge issues: Git’s automated code merging capabilities eliminate the need for manual code merge, which is prone to errors.</p><p>3. DevOps Ready: Organizations may embrace DevOps with ease because to Git’s simple integration with the wide variety of DevOps technologies available on the market.</p><p>4. Attract new talent: Providing new mainframe developers the environment which is familiar to them</p><p><strong>Chapter 5 : Endevor Bridge for Git — How does it work</strong></p><p>The Endevor Bridge for Git is included in the <strong>Endevor Integrations for Enterprise DevOps</strong> package. It connects Git to Endevor through an enterprise Git server such as GitHub, Bitbucket, GitLab, or Azure DevOps.</p><p>The architecture diagram illustrates the typical setup of the Endevor–Git Bridge, with the following key components:</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/917/1*pJx3IXI2AWjZOp7wJAMFpg.png" /></figure><ol><li><strong>Git Server ↔ Endevor Bridge for Git</strong><br> Communication between the Git server and Endevor Bridge for Git enables synchronization with Endevor. This connection is configured in the application.yml file by providing the Git server URL and credentials.</li><li><strong>Endevor Bridge for Git ↔ Endevor</strong><br> The communication link between Endevor Bridge for Git and Endevor is established through <strong>Endevor Web Services</strong>, which run under Tomcat on the mainframe.</li><li><strong>Event-Based Sync-Back</strong><br> Near real-time synchronization of changes from Endevor can be achieved using event-based sync-back. Administrators can configure a webhook in the Mainframe Webhook Server to monitor specific Endevor inventories connected to Endevor Bridge for Git mappings.</li><li><strong>Webhooks to External Systems</strong><br> Webhooks can also be configured to send Endevor Bridge for Git event notifications to external endpoints — for example, reporting successful synchronizations in a mapping to a Jenkins pipeline.</li></ol><p><strong>Chapter 6 : Endevor Team Build</strong></p><p>Endevor Team Build is a build engine that enables building and testing of Endevor projects on z/OS. It is composed of the following components:</p><ol><li><strong>Syncz</strong> — Allows users to edit files locally and then synchronize them with the mainframe.</li><li><strong>Exportz</strong> — Exports an Endevor location and generates project build files.</li><li><strong>Bldz</strong> — Handles the project build process.</li><li><strong>Publishz</strong> — Publishes artifacts from z/OS to repositories such as JFrog Artifactory or Nexus Artifactory.</li></ol><p><strong>Conclusion</strong></p><p>As organizations continue modernizing their software development practices, many look to harness the advanced capabilities of modern SCMs, particularly in supporting parallel development. However, a complete migration from mainframe SCMs to distributed systems is often seen as both risky and costly. Broadcom’s Endevor-Git bridge provides a practical solution, enabling mainframe teams to integrate Git and embrace DevOps practices while preserving their established core SCM processes.</p><p><em>References :</em></p><p><em>[1] </em><a href="https://www.zdnet.com/article/programming-languages-how-much-cobol-code-is-out-there-the-answer-might-surprise-you/"><em>https://www.zdnet.com/article/programming-languages-how-much-cobol-code-is-out-there-the-answer-might-surprise-you/</em></a></p><p><em>[2] </em><a href="https://techdocs.broadcom.com/us/en/ca-mainframe-software/devops/endevor-bridge-for-git/2-0.html"><em>https://techdocs.broadcom.com/us/en/ca-mainframe-software/devops/endevor-bridge-for-git/2-0.html</em></a></p><p><em>Disclaimer:</em></p><ol><li><em>The views and opinions expressed in this blog are solely those of the author and do not necessarily reflect the official policy or position of the author’s employer or any other organization. The information provided is based on personal experience and research and is intended for informational purposes only.</em></li><li><em>Blog is written by me , but enhanced by LLM :)</em></li></ol><img src="https://medium.com/_/stat?event=post.clientViewed&referrerSource=full_rss&postId=5f447fd0e217" width="1" height="1" alt="">]]></content:encoded>
        </item>
        <item>
            <title><![CDATA[AI-Driven Mainframe Transformation: A Look at Watsonx Code Assistant for Z]]></title>
            <link>https://medium.com/elevate-tech/ai-driven-mainframe-transformation-a-look-at-watsonx-code-assistant-for-z-a38b4a6812a0?source=rss-db0a0da6d449------2</link>
            <guid isPermaLink="false">https://medium.com/p/a38b4a6812a0</guid>
            <category><![CDATA[code]]></category>
            <category><![CDATA[ai]]></category>
            <category><![CDATA[mainframe]]></category>
            <category><![CDATA[ibm]]></category>
            <dc:creator><![CDATA[Vikas Pujar]]></dc:creator>
            <pubDate>Sat, 20 Sep 2025 07:47:12 GMT</pubDate>
            <atom:updated>2025-09-22T17:40:01.253Z</atom:updated>
            <content:encoded><![CDATA[<h3>AI-Driven Mainframe Transformation: A Look at watsonx Code Assistant for Z</h3><p><strong>Chapter 1 : Introduction</strong></p><p>In April 2024, Mainframe celebrated its 60th birthday, while Cobol celebrated its 65th. It is incredible to witness a technology or language survive, or should say thrive, for more than six decades. However, one cannot deny that it is difficult to find or attract new talent to work on Cobol, whereas there is a larger talent pool that knows Java; also, clients face challenges in maintaining and managing large monolithic Cobol applications.</p><p>Customers who perceive Mainframe as their strategic platform but want to address skill shortage, time to market, and complex monolithic application architecture can benefit from IBM’s new solution, watsonx code assistant for z.</p><p><strong>Chapter 2 : watsonx Code Assistant for Z Overview</strong></p><p>IBM released the Cobol to Java code transformation solution IBM watsonx Code Assistant for Z (WCA4Z henceforth) powered by GenAI in 2023. WCA4Z is an AI-assisted solution allowing enterprises to use generative AI and automated tooling to expedite mainframe application modernization while retaining IBM z’s performance, security, and resiliency.</p><p>The WCA4Z has five ( plus code explanation) features as shown in image.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*rE4DnwhWtvkRjZsmvcWVtw.jpeg" /><figcaption>WCA4Z Features</figcaption></figure><p>The functionality of each feature is listed below :</p><p>1. Understand: It is said that ‘the journey to modernization begins with understanding.’ As the name suggests, this feature enables users to understand the application to be modernized. The Z Understand tool is leveraged to discover the mainframe application code and dependencies.</p><p>2. Refactor: The <em>Refactor</em> feature allows users to select a snippet of code to be transformed from one or more programs and <em>Refactor</em> feature does a scan of components to pick the dependents within programs (ex. variables, nested paragraphs, etc.) and create a new Cobol program that can be used for transformation.</p><p>3. Optimize : This feature reviews the code + runtime logs to provide recommendation to improve performance of cobol code.</p><p>4. Transform: Here output of <em>Refactor</em> is fed to LLM (Granite models) for it to transform previously refactored Cobol code to Java.</p><p>5. Validate : The testing of previously transformed Java is performed in <em>Validate</em> ensuring transformed Java code works as expected.</p><p>6. Code Explanation : This feature takes code as input and provides simple explanation of code functionality, this step leverages IBMs LLM GenAI model Granite.</p><p>Over the past year, the WCA4Z tool and its underlying models have evolved. Initially, WCA4Z was available as both an on-premises and SaaS solution, using GenAI (LLM) models with code metadata stored in the cloud. In recent versions, all WCA4Z features can now be run entirely on-premises (x86).</p><p><strong>Chapter 3 : Drivers for usages of WCA4Z</strong></p><p>Here are some of the drivers for customers to consider usage of WCA4Z.</p><p>1. Skill shortage: Address skills<em> gaps by attracting new talent with modern tools</em> and <em>boost</em> productivity with automation and AI.</p><p>2. Time to market: Align with industry-standard DevOps approaches to improve time to market.</p><p>3. Simplify architecture: Transform monolith to manageable smaller components and utilize z quality of service with language interoperability.</p><p><strong>Chapter 4 : WCA4Z transformation approach</strong></p><p>WCA4Z modernization approach is to minimize risk with an iterative framework with selectively modernizing legacy apps based on business needs. This approach allows maintaining flexibility to leverage Cobol and Java with complete interoperability. The below image depicts the approach where services from large monolithic Cobol code is extracted and transformed to Java with interoperability between them.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/342/1*l9R5051dHKshIPKWOrxjDg.png" /><figcaption>Approach — Number of lines of code are just for illustrative purposes</figcaption></figure><p><strong>Chapter 5 : Conclusion</strong></p><p>WCA4Z is a GenAI-powered solution designed to streamline and accelerate the modernization of mainframe applications. It offers end-to-end support for the development lifecycle, including automated code refactoring, performance optimization, and Cobol-to-Java transformation. WCA4Z enhances developer productivity, reduces modernization risks, and allows organizations to adopt a faster, more cost-effective transformation process while maintaining mainframe performance standards.</p><p>References</p><p>[1] <a href="https://www.ibm.com/products/watsonx-code-assistant-z">https://www.ibm.com/products/watsonx-code-assistant-z</a></p><p>[2] <a href="https://techchannel.com/modernization/ibm-introduces-watsonx-code-assistant-for-z/">https://techchannel.com/modernization/ibm-introduces-watsonx-code-assistant-for-z/</a></p><p><em>Disclaimer:</em></p><ol><li><em>The views and opinions expressed in this blog are solely those of the author and do not necessarily reflect the official policy or position of the author’s employer or any other organization. The information provided is based on personal experience and research and is intended for informational purposes only.</em></li><li><em>Blog is written by me , but enhanced by LLM :)</em></li></ol><img src="https://medium.com/_/stat?event=post.clientViewed&referrerSource=full_rss&postId=a38b4a6812a0" width="1" height="1" alt=""><hr><p><a href="https://medium.com/elevate-tech/ai-driven-mainframe-transformation-a-look-at-watsonx-code-assistant-for-z-a38b4a6812a0">AI-Driven Mainframe Transformation: A Look at Watsonx Code Assistant for Z</a> was originally published in <a href="https://medium.com/elevate-tech">Elevate Tech</a> on Medium, where people are continuing the conversation by highlighting and responding to this story.</p>]]></content:encoded>
        </item>
        <item>
            <title><![CDATA[IBM z17: Ushering in the Next Era of AI-Powered Mainframes]]></title>
            <link>https://medium.com/elevate-tech/chapter-1-introduction-08eb6737a117?source=rss-db0a0da6d449------2</link>
            <guid isPermaLink="false">https://medium.com/p/08eb6737a117</guid>
            <category><![CDATA[ibm]]></category>
            <category><![CDATA[ai]]></category>
            <category><![CDATA[llm]]></category>
            <category><![CDATA[software-development]]></category>
            <dc:creator><![CDATA[Vikas Pujar]]></dc:creator>
            <pubDate>Mon, 15 Sep 2025 08:45:55 GMT</pubDate>
            <atom:updated>2025-09-24T10:45:11.049Z</atom:updated>
            <content:encoded><![CDATA[<p><strong>Chapter 1 : Introduction</strong></p><p>Enterprises are continuously being confronted with rapidly evolving demands around real-time AI inference, hybrid cloud integration, quantum-safe security, and resilient operations. Nearly five months since its release, IBM’s new mainframe z17 with integrated AI boosts innovation, supports new workloads, and increases productivity in a secure, reliable, and resilient environment tailored to the enterprise’s needs.</p><p><strong>Chapter 2 : How IBM z17 is different compared to its predecessors</strong></p><p>IBM refreshes its z-series mainframe hardware every 2–3 years, and in April 2025, it announced the latest iteration: the IBM z17. In addition to the usual upgrades in memory, storage, and performance, the z17 introduces two major hardware innovations that set it apart: the new Spyre Accelerator and the upgraded Telum II processor. Spyre is designed to enable organizations to run generative AI (GenAI) workloads natively on the mainframe, while Telum II powers large-scale core AI inference tasks with significantly enhanced performance. The Spyre chip, to be available later (Dec 2025 or early 2026), can run leading industry large language models (LLMs) like GPT, Llama, Claude, Gemini, etc., along with IBM Granite LLMs on the platform.</p><p>Both IBM Telum II and the Spyre Accelerator are engineered to support a broader and more complex range of AI models, particularly for use cases involving the ensemble AI method. Ensemble AI combines the outputs of multiple models to enhance the overall accuracy and performance of predictions, outperforming what individual models can achieve alone. Take insurance claims fraud detection as an example , traditional core AI can be used to generate an initial risk assessment based on structured data such as claim history, transaction patterns, and customer profiles. When this is combined with insights from LLMs which can analyze unstructured data like claim narratives, medical notes, or adjuster comments — the ensemble approach significantly improves both performance and accuracy. This combination enables deeper, more contextual understanding of potential fraud cases.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*Dmb_lfL6KdpHJZuCiA5epw.jpeg" /><figcaption>Image Source : IBM</figcaption></figure><p><strong>Chapter 3 : Key Features of IBM z17 (non-exhaustive)</strong></p><p><strong>1.</strong> <strong>Enhanced Processor and Memory</strong></p><p>IBM z17 Supports up to 64TB of system memory as compared to maximum 40 TB in z16, enabling larger in-memory workloads. IBM claims up to 20% more performance per core and up to 25% more system capacity compared to z16. IBM z17 uses and enhanced IBM Telum II processor, offering higher performance per core and 40 percent growth in cache.</p><p><strong>2.</strong> <strong>AI and On-Chip Acceleration for Transactions AI workloads</strong></p><p>IBM z17 expands AI capabilities with improved Telum II on-chip AI inferencing, supporting complex models and faster inferencing, enabling broader AI workloads directly on the mainframe, 50% more AI inference operations per day compared to z16.</p><p><strong>3.</strong> <strong>Expanded Quantum-Safe Security</strong></p><p>IBM z17 expanded the Quantum-safe security, with additional cryptographic algorithms and improved key management, future proofing against quantum threats.</p><p><strong>4.</strong> <strong>System Resiliency , Availability and Serviceability (RAS)</strong></p><p>IBM z17 further enhances the RAS features of IBM z machines, including improved predictive failure analysis and faster recovery from hardware faults and even higher system availability.</p><p><strong>5. Cloud and Hybrid Integration</strong></p><p>While IBM z16 enhanced support for hybrid cloud, containerization, and DevOps, IBM z17 builds on this with deeper integration for hybrid cloud, improved support for containerized workloads, and new APIs for automation and orchestration, making it easier to integrate with public and private clouds.</p><p><strong>6.</strong> <strong>Software Ecosystem</strong></p><p>IBM z17 comes with support for the latest z/OS versions 3.1 &amp; 2.5 and new software features, including expanded AI, security, and cloud-native capabilities. IBM z17 will be supporting the newer version of zOS 3.2 expected to be released by September’2025.</p><p><strong>7.</strong> <strong>Bring or Build AI models on z17</strong></p><p>With the ONNX open interface, organizations can seamlessly bring their existing machine learning models to the IBM Z platform, allowing them to run natively and efficiently with AI-optimized performance of the Telum II and Spyre chips. For those looking to build models directly on the platform, IBM Z also offers native support for developing and deploying AI workloads within the mainframe environment.</p><p>With its advanced features, the IBM z17 is purpose-built to handle the growing volume and complexity of mainframe workloads and other real-time operations that demand extreme scalability, high availability (up to 99.999999%), and robust throughput and resiliency.</p><p><strong>Conclusion:</strong></p><p>Earlier generations of mainframes (z15, z14, z13 etc) primarily focused on running core mainframe workloads and enabling hybrid cloud integration. With the launch of the z16 in 2022, powered by the Telum I chip, IBM pivoted to bringing AI directly onto the mainframe. The newly introduced IBM z17 is not just an incremental upgrade — it represents a true transformational leap by embedding AI acceleration, inference, generative capabilities, robust security, and hybrid-cloud support straight into the mainframe core, the z17 positions itself as the platform of choice for mission-critical, data-intensive AI workloads such as real-time fraud detection, risk scoring, and secure generative workflows. The Spyre chip on the z17 is a game changer, enabling large language models (LLMs) to run natively on the mainframe. It is a powerful statement that mainframes remain central in modern enterprise IT — reinvented for AI, deeply secure, and hybrid-ready.</p><p>References</p><p>[1] <a href="https://www.ibm.com/products/z17">https://www.ibm.com/products/z17</a></p><p>[2] <a href="https://www.redbooks.ibm.com/feature/z17">https://www.redbooks.ibm.com/feature/z17</a></p><p>[3] <a href="https://www.networkworld.com/article/3957317/ibm-z17-mainframe-gains-ai-accelerator-massive-workload-support.html">https://www.networkworld.com/article/3957317/ibm-z17-mainframe-gains-ai-accelerator-massive-workload-support.html</a></p><p><em>Disclaimer:</em></p><ol><li><em>The views and opinions expressed in this blog are solely those of the author and do not necessarily reflect the official policy or position of the author’s employer or any other organization. The information provided is based on personal experience and research and is intended for informational purposes only.</em></li><li><em>Blog is written by me , but enhanced by LLM :)</em></li></ol><p>#ElevateTech</p><img src="https://medium.com/_/stat?event=post.clientViewed&referrerSource=full_rss&postId=08eb6737a117" width="1" height="1" alt=""><hr><p><a href="https://medium.com/elevate-tech/chapter-1-introduction-08eb6737a117">IBM z17: Ushering in the Next Era of AI-Powered Mainframes</a> was originally published in <a href="https://medium.com/elevate-tech">Elevate Tech</a> on Medium, where people are continuing the conversation by highlighting and responding to this story.</p>]]></content:encoded>
        </item>
    </channel>
</rss>