<?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 Iftach Schonbaum on Medium]]></title>
        <description><![CDATA[Stories by Iftach Schonbaum on Medium]]></description>
        <link>https://medium.com/@iftachsc?source=rss-c51152b94193------2</link>
        <image>
            <url>https://cdn-images-1.medium.com/fit/c/150/150/2*mp7ryEL_-ropERVYjglMtA.png</url>
            <title>Stories by Iftach Schonbaum on Medium</title>
            <link>https://medium.com/@iftachsc?source=rss-c51152b94193------2</link>
        </image>
        <generator>Medium</generator>
        <lastBuildDate>Wed, 18 Nov 2026 18:35:54 GMT</lastBuildDate>
        <atom:link href="https://medium.com/@iftachsc/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[CloudZone wins Amazon Web Services 2021 Israel Partner Award]]></title>
            <link>https://medium.com/cloudzone/cloudzone-wins-amazon-web-services-2021-israel-partner-award-e82de023a9b6?source=rss-c51152b94193------2</link>
            <guid isPermaLink="false">https://medium.com/p/e82de023a9b6</guid>
            <category><![CDATA[press-release]]></category>
            <category><![CDATA[amazon-web-services]]></category>
            <category><![CDATA[winner]]></category>
            <category><![CDATA[cloud-computing]]></category>
            <category><![CDATA[aws-partner-network]]></category>
            <dc:creator><![CDATA[Iftach Schonbaum]]></dc:creator>
            <pubDate>Sun, 05 Dec 2021 12:04:14 GMT</pubDate>
            <atom:updated>2021-12-05T12:04:14.932Z</atom:updated>
            <content:encoded><![CDATA[<figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*yHJ4XZJE6EAY24XzznXpVg.png" /></figure><p>CloudZone has won an AWS Partner Award recognizing leaders in Israel playing a key role helping customers drive innovation and build solutions on the AWS Cloud.</p><p>CloudZone won the AWS Consulting Partner of the Year award. The award recognizes an AWS Consulting Partner who has leveraged the best of AWS Services and continued to expand and grow with AWS during 2021.</p><p>The AWS Partner Awards recognize a wide range of Partners, whose business models have embraced specialization and collaboration over the past year. The AWS Partner Awards also recognize AWS Partners whose business models continue to evolve and thrive on the AWS Cloud as they work with customers, enabling them to build and co-sell with AWS.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*6rSmJ4BIi86GCcsew5J79w.png" /></figure><p>The AWS Partner Network (APN) is the global community of businesses using AWS to build solutions and services for customers. AWS helps partners build, market, and sell their offerings by providing valuable business, marketing, and technical support on top of our proven, global, and secure infrastructure. AWS partners are uniquely positioned to help customers take full advantage of all that AWS has to offer and accelerate their journey to the cloud. AWS has millions of active customers and a dynamic community of over 100,000 partners globally.</p><p>“AWS Partners add value to customers across a wide range of industries around the world and are critical to their success. We are pleased to recognize a variety of partners through the APN Partner Awards EMEA 2021. I look forward to working together to a great 2022,” said Antonio Alonso Lopez, EMEA Director Partner Success at AWS.</p><img src="https://medium.com/_/stat?event=post.clientViewed&referrerSource=full_rss&postId=e82de023a9b6" width="1" height="1" alt=""><hr><p><a href="https://medium.com/cloudzone/cloudzone-wins-amazon-web-services-2021-israel-partner-award-e82de023a9b6">CloudZone wins Amazon Web Services 2021 Israel Partner Award</a> was originally published in <a href="https://medium.com/cloudzone">CloudZone</a> on Medium, where people are continuing the conversation by highlighting and responding to this story.</p>]]></content:encoded>
        </item>
        <item>
            <title><![CDATA[CloudZone invests in Portugal and brings its cost-free cloud consulting services]]></title>
            <link>https://medium.com/cloudzone/cloudzone-invests-in-portugal-and-brings-its-cost-free-cloud-consulting-services-2f8df5d9d122?source=rss-c51152b94193------2</link>
            <guid isPermaLink="false">https://medium.com/p/2f8df5d9d122</guid>
            <category><![CDATA[aws-partner-network]]></category>
            <category><![CDATA[press-release]]></category>
            <category><![CDATA[cloud-computing]]></category>
            <category><![CDATA[aws]]></category>
            <category><![CDATA[multi-cloud]]></category>
            <dc:creator><![CDATA[Iftach Schonbaum]]></dc:creator>
            <pubDate>Mon, 04 Oct 2021 13:31:23 GMT</pubDate>
            <atom:updated>2021-10-04T13:31:23.811Z</atom:updated>
            <content:encoded><![CDATA[<figure><img alt="" src="https://cdn-images-1.medium.com/max/720/0*Pm5Hub0qSuY03vPf.jpg" /></figure><p><a href="https://cloudzone.pt/">CloudZone</a>, an Israeli founded leading top-tier, multi-cloud Managed Service Provider (MSP), announces a strategic investment in the Portuguese market and opens its first local office in Porto, Portugal. CloudZone joins forces with Flipkick, a recognized Portuguese cloud consulting and professional services company, to form CloudZone Portugal.</p><p>Cloudzone Portugal will accelerate organizations’ cloud journey in the Iberian Peninsula. Nuno Tavares, CEO and founder of Flipkick, will act as CloudZone Portugal country manager.</p><p>For over 8 years, CloudZone has successfully accelerated hundreds of global businesses across all segments, sectors, and verticals. It brings vast and diverse cloud knowledge, including best practices and cutting-edge solutions in technology domains such as Migrations, DevOps, Cloud Native, Security, FinOps, Big Data &amp; Analytics, Machine Learning &amp; Artificial Intelligence (ML&amp;AI), Internet of Things (IoT), and others.</p><p>Being a subsidiary of Matrix, the largest Information Technology company in Israel, CloudZone provides a one-stop-shop with a wide range of services including security, software development, outsourcing and solutions tailored for the Finance, Healthcare, Retail, Defense, Hi-Tech and Public Sectors.</p><blockquote>“Unlike other consulting partners CloudZone leverages its strong alliance with the Cloud vendors to accelerate its customers by providing cost-free and unmetered consulting, customer support, FinOps (cost-optimization) and cloud reporting &amp; governance system, in addition to value added services including Professional Services, 24x7 NOC and IT training programs.</blockquote><blockquote>Our cost-free service model and proactive management, aligned with customer priorities, helps businesses to migrate, evolve and innovate faster, to well-architect their cloud solution, to solve performance and complexity issues, to cut cloud costs wastage, thus improving agility and overall business results. After reaching maturity to accelerate hundreds of customers efficiently and successfully, CloudZone now strives to replicate its service model to new destinations and to bring those benefits to the Iberian region too”, said Erez Tourjeman, CloudZone Regional Manager, EMEA.</blockquote><blockquote>Nuno Tavares, CloudZone Portugal Country Manager adds “This is another sign that Portuguese technology companies continue to attract the attention of major international players. CloudZone long and successful delivery record brings a much-needed and dramatic boost to the local market supporting its outstanding up-trend”.</blockquote><p>CloudZone started its strategic expansion into Europe this year, starting from the Iberian Peninsula, where demand for experienced, agile, personalized, and affordable cloud consulting services was identified. It already delivers to a few large Portuguese customers.</p><blockquote>José Silva, Swogo’s CTO says “Having CloudZone’s team expertise at our service at no additional costs while optionally being able to use their value-added services to accelerate us proved to be no-brainer for our business”</blockquote><h3><strong>CloudZone is AWS Premier MSP Partner</strong></h3><p>CloudZone provides migration and deployment services to businesses of all sizes and is an Amazon Web Services (AWS) Premier Consulting Partner with expertise in multiple cloud technologies including public, private and hybrid environments.</p><p>In 2020, CloudZone was awarded Amazon Web Services <a href="https://cloudzone.io/apn-devops-partner-of-the-year/">APN DevOps Partner of the Year award</a>. As a certified <a href="https://partners.amazonaws.com/partners/001E000000YM87XIAT/CloudZone">AWS Premier MSP partner</a>, with verified competencies including Migration, DevOps, Security, Well-Architected, Public Sector, Windows Workloads, VMware on AWS and more. CloudZone efficiently helps its customers technically and commercially through Amazon Web Services Partner Acceleration Programs.</p><p>João Tedim, Partner success manager at AWS Portugal, points out “We welcome CloudZone and its Agile &amp; Proactive delivery experience to Portugal. Together, the companies are providing a complete cloud services and cloud management portfolio that will give customers of all sizes fast, flexible access to the cloud. Building on the existing relationship with CloudZone, now in Portugal, reinforces our commitment to the country, to accelerate cloud computing. Together, we will continue the work we have been doing, empowering any person or industry, sharing cloud computing knowledge, accelerating innovation and the Portuguese economy.”</p><img src="https://medium.com/_/stat?event=post.clientViewed&referrerSource=full_rss&postId=2f8df5d9d122" width="1" height="1" alt=""><hr><p><a href="https://medium.com/cloudzone/cloudzone-invests-in-portugal-and-brings-its-cost-free-cloud-consulting-services-2f8df5d9d122">CloudZone invests in Portugal and brings its cost-free cloud consulting services</a> was originally published in <a href="https://medium.com/cloudzone">CloudZone</a> on Medium, where people are continuing the conversation by highlighting and responding to this story.</p>]]></content:encoded>
        </item>
        <item>
            <title><![CDATA[Cloudzone awarded AWS DevOps Partner of the Year]]></title>
            <link>https://medium.com/cloudzone/cloudzone-awarded-aws-devops-partner-of-the-year-19a33b57eced?source=rss-c51152b94193------2</link>
            <guid isPermaLink="false">https://medium.com/p/19a33b57eced</guid>
            <category><![CDATA[aws-partner]]></category>
            <category><![CDATA[devops-practice]]></category>
            <category><![CDATA[aws-partner-network]]></category>
            <category><![CDATA[aws]]></category>
            <category><![CDATA[press-release]]></category>
            <dc:creator><![CDATA[Iftach Schonbaum]]></dc:creator>
            <pubDate>Mon, 14 Jun 2021 10:35:04 GMT</pubDate>
            <atom:updated>2021-06-14T10:35:04.698Z</atom:updated>
            <content:encoded><![CDATA[<figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*B82qqe4uz9O5-2mO4gIoZg.png" /></figure><p>CloudZone, an Amazon Web Services (AWS) Premier Consulting Partner since 2014, announced today that it has been awarded an APN DevOps Partner of the Year Award for 2020. With over 10 years of experience in cloud consulting, delivery and support, CloudZone provides its customers with a complete service package that includes top-notch professional services, managed services, <a href="https://cloudzone.io/finops/">FinOps services</a> and consulting. The APN Partner Awards recognize members of the AWS Partner Network (APN) who are leaders in the channel and play a key role in helping customers to drive innovation and build solutions on AWS.</p><p>CloudZone’s extensive experience with AWS, and its track record of winning customers and delivering highly complex projects, position the company as a trusted advisor among customers. Having achieved <a href="https://cloudzone.io/aws-devops-competency/">AWS DevOps Competency</a> in January 2019, CloudZone has grown rapidly, achieved great success around <a href="https://cloudzone.io/devops/">DevOps expertise</a> and made a significant contribution to the AWS business. This award recognizes that the company is ‘customer obsessed’ and delivers results and innovation in DevOps practices.</p><p>The APN Award is also testament to CloudZone’s ability to significantly accelerate its customers’ businesses, by allowing them to focus on tasks that improve their growth and bottom line. As a CNCF Silver Member and a Kubernetes Certified Service provider, CloudZone continuously develops and polishes tools for automation, provisioning and monitoring of AWS infrastructure according to customer requirements and building CI/CD pipelines to maximize development efficiency.</p><p>CloudZone is always on the lookout for new ways to automate tasks and create tools that help its customers’ operations work more efficiently. The company also takes great care to refine its own monitoring system, so that it can detect negative trends and take remedial action before problems arise. CloudZone’s DevOps service includes all the benefits of operational efficiency, enhanced insight, auto-scale on demand, simpler multi-zone deployments, continuous integration and overall improved bottom line.</p><p>“We are honored and proud to receive this award, which recognizes our commitment to AWS and to continuous innovation in DevOps, while enabling our customers to focus their business goals,” said Adi Heinisch, General Manager, CloudZone. “As AWS DevOps Competency holders and as an AWS Premier Consulting Partner, we are able to deliver outstanding DevOps services that dramatically accelerate initiatives on AWS, and maximizes our customers’ business impacts. We look forward to contributing to our customers’ success further as we continue to collaborate with AWS.”</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*57riKJX8hyM9afOnHtGeCQ.png" /><figcaption>Photo By : Omer Barash</figcaption></figure><p><strong>About Our Cloud Services</strong></p><p>CloudZone helps you leverage the power of the Cloud, so that you can focus on your core business strategies. As a <a href="https://cloudzone.io/">multi-cloud service provider</a>, we help customers to take advantage of the broad set of global compute, storage, data, analytics, application, and deployment services. Our goal is to help organizations move faster, lower their IT costs, and scale their applications.</p><p>We ensure that our customers achieve cost and operational efficiency and adopt best practices for resource consumption management, cost optimization, and governance. Delivering an additional layer of proactive management and 24/7 support, we give our customers the assurance of knowing that their infrastructure is fully and effectively monitored.</p><p><a href="http://cloudzone.io/">Cloud Services By CloudZone - Multi Cloud Solution Provider</a></p><img src="https://medium.com/_/stat?event=post.clientViewed&referrerSource=full_rss&postId=19a33b57eced" width="1" height="1" alt=""><hr><p><a href="https://medium.com/cloudzone/cloudzone-awarded-aws-devops-partner-of-the-year-19a33b57eced">Cloudzone awarded AWS DevOps Partner of the Year</a> was originally published in <a href="https://medium.com/cloudzone">CloudZone</a> on Medium, where people are continuing the conversation by highlighting and responding to this story.</p>]]></content:encoded>
        </item>
        <item>
            <title><![CDATA[Get More Value From Snowflake With Upsolver]]></title>
            <link>https://medium.com/cloudzone/get-more-value-from-snowflake-with-upsolver-620dc19a3813?source=rss-c51152b94193------2</link>
            <guid isPermaLink="false">https://medium.com/p/620dc19a3813</guid>
            <category><![CDATA[data-lake]]></category>
            <category><![CDATA[webinar]]></category>
            <category><![CDATA[cloud-data-warehouse]]></category>
            <category><![CDATA[big-data-analytics]]></category>
            <category><![CDATA[stream-processing]]></category>
            <dc:creator><![CDATA[Iftach Schonbaum]]></dc:creator>
            <pubDate>Mon, 14 Jun 2021 08:38:38 GMT</pubDate>
            <atom:updated>2021-06-14T09:00:01.889Z</atom:updated>
            <content:encoded><![CDATA[<figure><img alt="" src="https://cdn-images-1.medium.com/max/747/1*Kip-7togGGpd-VygDVIWPA.png" /></figure><p>Learn how Combining Snowflake with Upsolver’s Cloud Lakehouse engine empowers organizations to achieve the improved cost and flexibility advantages of a data lake, with the ease of use of a data warehouse.</p><p>Deciding on the right data platform for your business is a crucial component of your big data strategy. The right data platform can deliver a powerful return on investment, transforming your business by giving you keener data-driven insights.</p><p>Snowflake is a mature, robust Cloud-based data platform giant used by thousands of happy customers. Combining Snowflake with Upsolver’s Cloud-native lakehouse engine empowers your organization to achieve the improved cost and flexibility advantages of a data lake, with the ease of use of a data warehouse.</p><iframe src="https://cdn.embedly.com/widgets/media.html?src=https%3A%2F%2Fwww.youtube.com%2Fembed%2FjrAWOjg6j6g%3Ffeature%3Doembed&amp;display_name=YouTube&amp;url=https%3A%2F%2Fwww.youtube.com%2Fwatch%3Fv%3DjrAWOjg6j6g&amp;image=https%3A%2F%2Fi.ytimg.com%2Fvi%2FjrAWOjg6j6g%2Fhqdefault.jpg&amp;key=a19fcc184b9711e1b4764040d3dc5c07&amp;type=text%2Fhtml&amp;schema=youtube" width="854" height="480" frameborder="0" scrolling="no"><a href="https://medium.com/media/f203732c9a00cd13ebd23dcb16282c43/href">https://medium.com/media/f203732c9a00cd13ebd23dcb16282c43/href</a></iframe><img src="https://medium.com/_/stat?event=post.clientViewed&referrerSource=full_rss&postId=620dc19a3813" width="1" height="1" alt=""><hr><p><a href="https://medium.com/cloudzone/get-more-value-from-snowflake-with-upsolver-620dc19a3813">Get More Value From Snowflake With Upsolver</a> was originally published in <a href="https://medium.com/cloudzone">CloudZone</a> on Medium, where people are continuing the conversation by highlighting and responding to this story.</p>]]></content:encoded>
        </item>
        <item>
            <title><![CDATA[Kubeflow and ML Automation: Part 1]]></title>
            <link>https://medium.com/cloudzone/kubeflow-and-ml-automation-part-1-fcb8e8040ac6?source=rss-c51152b94193------2</link>
            <guid isPermaLink="false">https://medium.com/p/fcb8e8040ac6</guid>
            <category><![CDATA[machine-learning]]></category>
            <category><![CDATA[cloud-native]]></category>
            <category><![CDATA[ml-engineering]]></category>
            <category><![CDATA[kubernetes]]></category>
            <category><![CDATA[mlops]]></category>
            <dc:creator><![CDATA[Iftach Schonbaum]]></dc:creator>
            <pubDate>Wed, 14 Apr 2021 12:31:30 GMT</pubDate>
            <atom:updated>2021-04-14T12:31:30.674Z</atom:updated>
            <content:encoded><![CDATA[<figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*l_G1mrhz2c3mk_OK8bZ74g.png" /></figure><p>With the growing maturity of the machine learning (ML) ecosystem and the deeper integration of ML algorithms into production software, managing the development, testing, and deployment of ML models has become a complex task.</p><p>Training deep neural-network models in a cloud environment requires a highly customized system that links together different components, such as compute, storage, and networking, allowing you to manage and orchestrate an ML pipeline in a consistent way. To create a functional ML pipeline, ML practitioners need to be able to set up an ML development environment, provision and scale compute power for training their models, create the models’ API, serve the models, and manage their lifecycle. But handling these tasks manually is an error-prone and time-consuming process. Moreover, not all ML practitioners have the DevOps expertise required to go from development to production and manage the AI/ML pipeline.</p><p>Automation can help address many of these challenges and is an integral part of the MLOps methodology, which aims to streamline ML workflows throughout the application lifecycle. In general, automation can provide the following benefits for the ML workflow in a distributed compute environment:</p><ul><li>Lifecycle management of ML training jobs, including automatic scaling</li><li>Composing, linking, and orchestrating different components of the ML pipeline</li><li>Ensuring ML jobs have high availability and fault tolerance via automatic health checking and recovery</li><li>Reproducibility of ML experiments and enabling iterative ML practices e.g. automatic retraining based on incoming data</li><li>Automatic provisioning of compute resources</li></ul><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/0*ctyL27HdkN5YRr4P.png" /><figcaption>MLOPS Pipeline (Image by Kaskada.com)</figcaption></figure><p>This blog post is Part 1 of the “Kubeflow and ML Automation” series, which describes how Kubeflow helps automate Machine Learning on Kubernetes.</p><p>In this post, we introduce readers to <a href="https://www.kubeflow.org/">Kubeflow</a>, an open-source Kubernetes-based tool for automating your ML workflow. We show how the Kubeflow Pipelines platform and Kubeflow components allow you to automate and manage different stages of the ML workflow, including data preparation, model experimentation, training, and deployment. We also discuss how Kubeflow can be used as a part of the cloud-based managed K8s service.</p><p>In Part 2, we’ll walk readers through some practical examples of using Kubeflow for model training, ML model optimization, serving, metadata retrieval and processing, and creating composable and reusable ML pipelines.</p><h3>Why ML Automation?</h3><p>The traditional process of ML research and development is based on multiple manual practices such as data pre-processing, model selection and testing, model optimization, and deployment. This process requires a lot of specialized knowledge in mathematics, statistics, and programming.</p><p>Although advanced ML practitioners are equipped with the knowledge necessary to develop ML models, the process of testing, deploying, and training ML models requires specialized expertise in compute and storage infrastructure, as well as networking, to serve and deploy the ML models. This complicates the process of deploying models to production and prevents many companies without specialized expertise in ML and DevOps from adopting AI/ML. Also, traditional ML processes lack reproducibility and repeatability and do not allow effective collaboration between different IT teams.</p><p>Automation, which is ubiquitous in computer programming and IT, is the natural solution to these ML challenges. It provides many benefits for companies seeking to adopt ML algorithms in their applications, including:</p><ul><li><strong>Faster TTM (time to market).</strong> ML automation allows you to streamline ML model training, testing, and deployment, which results in a faster transition from development to production.</li><li><strong>Enabling MLOps</strong>. Automation helps integrate various components of the ML workflow into a coherent pipeline that can be easily upgraded, tested, maintained, and deployed. Integrating automated testing, model builds, and deployments into the ML workflow aligns ML processes with existing CI/CD tools and approaches.</li><li><strong>Better collaboration</strong>. Reproducibility of automated ML experiments and automated metadata and artifact management leads to a clear understanding by different teams of the model development timeline. This ensures more efficient collaboration across teams and different projects.</li><li><strong>Reduction of human error</strong>. Subtle human errors can lead to a drastic deterioration in ML model performance, which is hard to debug due to the complexity of neural architecture and the “black-box” nature of model layers and parameters. Automation helps reduce human errors in ML models that are due to manual practices.</li><li><strong>Improved model accuracy and performance</strong>. <a href="https://cloud.google.com/automl">AutoML</a> algorithms can improve model accuracy and performance much faster than manual trial-and-error tuning, plus achieve better performance with real-world production data.</li></ul><p>ML automation provides numerous benefits for ML researchers and practitioners as well. Thanks to automated MLOps processes and the AutoML algorithms, they can develop, train, and optimize their model faster by focusing on the research part of their work, such as experimentation, and not having to worry as much about provisioning compute resources, implementing distributed training, and configuring training environments.</p><h3>Why Kubeflow Is the Answer to the Challenge of Automation</h3><p><a href="https://www.kubeflow.org/">Kubeflow</a> is an open-source ML platform designed to train and serve ML models on Kubernetes. The main purpose of Kubeflow is to enable MLOps — a methodology for the end-to-end management of ML workflows that facilitates fast model development, training, and rollout/upgrade cycles. To achieve this, Kubeflow leverages Kubernetes API resources and provides a set of tools to automate various stages of the ML workflow, from development and testing to deployment.</p><p>Kubeflow allows you to automate ML workflows in many different ways. Let’s discuss the most important of them.</p><h3>Automated Containerization of ML Code</h3><p>Training ML models in a distributed containerized environment like Kubernetes requires packaging ML code into containers configured with all the necessary executables, libraries, configuration files, and other data. Packaging containers whenever the model is updated takes time and may require a complex CI/CD pipeline. <a href="https://www.kubeflow.org/docs/components/fairing/fairing-overview/">Kubeflow Fairing </a>is a Kubeflow component that can automatically build container images from Jupyter Notebooks and Python files. Using Kubeflow Fairing, practitioners with less experience in Docker can easily run containers on Kubernetes.</p><h3>Automation of the Application Lifecycle in Kubernetes</h3><p>Distributed environments like Kubernetes clusters are highly volatile and dynamic, which makes it hard to ensure the uninterrupted operation of ML applications and training jobs. With hundreds of containers and pods running in one cluster, manual redeployment of failed applications and pods can be difficult, error-prone, and time-consuming.</p><p>Kufeblow leverages Kubernetes Control Plane and its operators to manage the lifecycle of training jobs. Kubeflow controllers such as TFOperator or PyTorch Operator interact with Kubernetes controllers and schedulers to perform automated health checks to ensure that the ML model is running as expected. In its turn, Kubernetes automatically restarts pods on failure and maintains the desired number of replicas in Deployments, StatefulSets, and Kubeflow custom API resources.</p><h3>Autoscaling with Kubeflow</h3><p>Autoscaling ensures the high availability of served ML models. The failure to dynamically scale ML inference servers based on inbound user requests can lead to downtimes and poor ML user experience. Kubeflow provides several tools for autoscaling served models on Kubernetes.</p><p>For example, autoscaling is supported for ML models deployed with the <a href="https://github.com/kubeflow/kfserving">KFServing framework</a>, which is a part of the Kubeflow default installation. Under the hood, KFServing uses <a href="https://knative.dev/docs/serving/autoscaling/">Knative </a>to autoscale deployments based on the average number of incoming requests per pod. If the concurrency target is set to three, for example, hundred concurrent requests will result in the spinning up of two new inference pods. The concurrency target can be easily set using the KFServing InferenceService custom resource. Similarly, flexible autoscaling is supported by <a href="https://github.com/SeldonIO/seldon-core">Seldon Core,</a> which is not part of Kubeflow but integrates well with it.</p><h3>Automated Distributed Training</h3><p>Distributed training is required for fast and scalable training of large ML models. It’s a natural solution in a distributed environment like Kubernetes where multiple nodes and CPUs/GPUs are available on-demand. However, it may be hard to manually configure the interaction and coordination between ML training jobs and workers, including the exchange and update of weights, computation of aggregate losses, etc.</p><p>Kubeflow ships with several tools that allow you to automate the distributed training of ML models on Kubernetes. For example, <a href="https://www.kubeflow.org/docs/components/training/tftraining/">TFJob</a> can be used to configure distributed training for TensorFlow models. The <a href="https://github.com/kubeflow/tf-operator">TFOperator</a> that manages these training jobs provides three abstractions that represent different agents in the distributed training: <strong>Chiefs</strong> are responsible for the training orchestration and model checkpointing, <strong>Parameter Servers</strong> perform weight updates and model loss calculations, and <strong>Workers</strong> run the training code. These abstractions can be used to implement both synchronous and asynchronous distributed training patterns supported by the TF distributed training modules.</p><p>To enable distributed training in your Kubeflow cluster, you can also use <a href="https://github.com/kubeflow/mpi-operator">MPI Operator,</a> which allows for allreduce-style distributed training of your ML models on Kubernetes. MPI Operator is the Kubeflow implementation of the Message Passing Interface (MPI), a multi-network protocol for efficient communication and coordination of nodes in compute clusters.</p><h3>Automation of Model Optimization</h3><p>ML model optimization is an important component of the ML workflow and aims to improve the performance and accuracy of trained ML models. Methods such as hyperparameter optimization and model architecture search have traditionally relied on repetitive trial and error experiments that take much time and are hard to test and reproduce. <a href="https://cloud.google.com/automl">AutoML</a> algorithms were developed to enable faster optimization of ML models for better accuracy and performance.</p><p>The <a href="https://github.com/kubeflow/katib">Kubeflow Katib</a> component provides AutoML tools for hyperparameter optimization and neural architecture search (NAS). Instead of testing hyperparameters manually, ML developers can use Katib to define the hyperparameter search space and let Katib perform the search using a specified AutoML algorithm. Katib supports algorithms such as <a href="https://arxiv.org/pdf/1012.2599.pdf">Bayesian optimization</a>, <a href="https://papers.nips.cc/paper/2011/file/86e8f7ab32cfd12577bc2619bc635690-Paper.pdf">Tree of Parzen Estimators</a>, and <a href="https://en.wikipedia.org/wiki/Hyperparameter_optimization#Random_search">Random Search</a>, among others.</p><p>This optimization can be performed on several hyperparameters in parallel. For example, you can simultaneously optimize the learning rate and regularization parameter. Also, Katib supports early-stopping algorithms to prevent model overfitting and NAS to select the optimal combination of a neural network’s layers and modules.</p><h3>Automation of ML Metadata and Log Management</h3><p>The conventional metadata and log management process used by ML practitioners typically involves the following steps:</p><ul><li>Generating artifacts (graphs, tables), logs, and metrics</li><li>Watching them while the training job is running</li><li>Recording the most important metrics for model performance analysis</li></ul><p>If used repeatedly over a prolonged period of time, this manual process can lead to losing track of past ML experiments. It’s difficult to monitor the history of the model and collect insights from the disparate and unmanaged logs and metadata generated by different experiments.</p><p><a href="https://www.kubeflow.org/docs/components/metadata/">Kubeflow Metadata</a> is a tool that helps you automate ML metadata management and address the above challenges. It ships with the Python SDK that lets ML practitioners record metadata directly from the ML model scripts. The SDK provides a number of useful functions to retrieve and organize the metadata on model training, datasets, metrics, and experiment runs and ship this data to Kubeflow Artifact Store for use by other components. Automatic retrieval of the metadata via Kubeflow enables better observability of ML experiments and lets you audit past experiments, performance metrics, datasets, and frameworks. A better understanding of past experiments enabled by Metadata can result in faster ML development cycles and better coordination across ML teams.</p><h3>Automating ML Workflow with Kubeflow Pipelines</h3><p>The Kubeflow automation tools we’ve discussed until now help automate specific parts of the ML workflow, such as training or hyperparameter optimization. The goal of <a href="https://www.kubeflow.org/docs/pipelines/overview/pipelines-overview/">Kubeflow Pipelines </a>is to automate the entire ML workflow and transform it into a composable ML Pipeline.</p><p>Kubeflow Pipelines consist of multiple components that represent various stages of the ML process in the form of a graph. For example, the pipeline can start with the data pre-processing job that consumes data from cloud storage and passes it to the training job. In turn, a training job component can be connected to the serving module that saves the trained model and launches the inference service to expose it to the Internet. This pipeline can be created using Pipeline DSL, saved as a separate file, and viewed as a graph in the Kubeflow dashboard.</p><p>Organizing the ML pipeline as a set of independent and connected components provides enormous benefits for AI/ML teams enabling the ML process in line with the MLOps methodology. Such an approach allows for modularity of the ML pipeline, similar to a microservices architecture. Each component of the pipeline can be developed, tested, and upgraded independently by different teams. This modularity leads to better reusability of ML components, as now, each of them can be used as components of other ML pipelines. Since components are abstracted and isolated from the underlying environment, they can be used in any pipeline that follows the same approach.</p><p>Also, Kubeflow pipelines enable easy repeatability and reproducibility of ML experiments. ML practitioners can run Kubeflow Pipelines knowing the execution order of different scripts and components, leading to improved coordination across teams and a better understanding of the ML workflow.</p><h3>Advantages of Using Kubeflow in the Cloud</h3><p>Running Kubeflow in the cloud provides additional benefits for companies seeking to automate their ML workflow and make it more efficient. Cloud providers like Google Cloud offer managed Kubernetes services and cloud-based ML platforms that can be integrated with the Kubeflow installation. Most importantly, when running Kubeflow in the cloud, you get access to highly performant on-demand CPUs/GPUs and storage.</p><p>Companies running Kubeflow in the cloud can take advantages of mature ecosystems of cloud tools including Big Data tools and databases, block storage, serverless cloud functions, monitoring, logging, tracing, auditing, security and add-on ML services that can be integrated into your own ML process as helpers or part of the pipeline.</p><p>For example, when running Kubeflow in Google Cloud, you can have access to Google Tensor Processing Units (TPUs), specialized hardware optimized for parameter computations and weights — updates typically made by the neural networks.</p><h3>Conclusion</h3><p>Kubeflow is one of the first tools to bring full automation of AI/ML pipelines in containerized and distributed environments like Kubernetes. It leverages Kubernetes API resources and orchestration services to ensure high availability and fault tolerance of containerized ML applications and provides its own set of tools to automate various parts of the ML workflow. Kubeflow’s feature set enables automation of ML model training, hyperparameter optimization, feature engineering, data pre-processing, model serving, and ML model containerization.</p><p>In addition to that, Kubeflow Pipelines help create composable ML workflows in line with the MLOps methodology. Pipelines makes ML workflows composable, reproducible, and extendable, dramatically improving the quality of ML collaboration across teams and enabling faster experimentation and better observability of the model training.</p><p>Built-in automation offered by Kubeflow makes it a great tool for companies looking for a fast and efficient way to deploy ML models to production. The platform can dramatically reduce time to market for ML products and facilitate efficient CI/CD processes in line with MLOps methodology.</p><p>If you want to learn more about using Kubeflow for the automation of ML on Kubernetes, you can read Part 2 of our series, which provides many practical examples of leveraging Kubeflow components to create efficient ML workflows.</p><img src="https://medium.com/_/stat?event=post.clientViewed&referrerSource=full_rss&postId=fcb8e8040ac6" width="1" height="1" alt=""><hr><p><a href="https://medium.com/cloudzone/kubeflow-and-ml-automation-part-1-fcb8e8040ac6">Kubeflow and ML Automation: Part 1</a> was originally published in <a href="https://medium.com/cloudzone">CloudZone</a> on Medium, where people are continuing the conversation by highlighting and responding to this story.</p>]]></content:encoded>
        </item>
        <item>
            <title><![CDATA[JenkinsX Deep Dive (Meetup)]]></title>
            <link>https://medium.com/cloudzone/jenkinsx-deep-dive-meetup-91871b2aa865?source=rss-c51152b94193------2</link>
            <guid isPermaLink="false">https://medium.com/p/91871b2aa865</guid>
            <category><![CDATA[jenkins-x]]></category>
            <category><![CDATA[continuous-deployment]]></category>
            <category><![CDATA[continuous-integration]]></category>
            <category><![CDATA[kubernetes]]></category>
            <category><![CDATA[webinar]]></category>
            <dc:creator><![CDATA[Iftach Schonbaum]]></dc:creator>
            <pubDate>Thu, 29 Oct 2020 13:07:22 GMT</pubDate>
            <atom:updated>2020-10-29T13:12:53.128Z</atom:updated>
            <content:encoded><![CDATA[<figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/0*FlddTNXviWiHcteD" /></figure><blockquote>JenkinsX is an open source opinionated way to do continuous delivery with Kubernetes, natively, without worrying too much about the underlying infrastructure.</blockquote><p>Watch our meetup on the subject, down below.</p><h4><strong>Agenda:</strong></h4><p><strong>Introduction</strong></p><ul><li>Hello Jenkins X, Bye bye Jenkins?</li><li>JenkinsX as a microservices app and its BOM.</li></ul><p><strong>Jenkins X</strong></p><ul><li>JX Boot</li><li>Applications &amp; Add-ons</li><li>Preview environments and promotion.</li><li>JenkinsX Pipelines &amp; TektonCD</li></ul><p><strong>Builders Build Packs</strong></p><ul><li>Short Overview on Lighthouse (ChatOps), Kaniko (Image build) &amp; Velero (K8s Backup)</li><li>JXUI, JX &amp; JXL</li></ul><p><strong>Case study — lesson learned when migrating to JenkinsX on AWS</strong></p><p><strong>Special Thanks</strong></p><p><strong>Road Map</strong></p><iframe src="https://cdn.embedly.com/widgets/media.html?src=https%3A%2F%2Fwww.youtube.com%2Fembed%2FxZ6b-aYw67U%3Ffeature%3Doembed&amp;display_name=YouTube&amp;url=https%3A%2F%2Fwww.youtube.com%2Fwatch%3Fv%3DxZ6b-aYw67U&amp;image=https%3A%2F%2Fi.ytimg.com%2Fvi%2FxZ6b-aYw67U%2Fhqdefault.jpg&amp;key=a19fcc184b9711e1b4764040d3dc5c07&amp;type=text%2Fhtml&amp;schema=youtube" width="854" height="480" frameborder="0" scrolling="no"><a href="https://medium.com/media/e2a861291c6c1fa9af410556eba7c8f7/href">https://medium.com/media/e2a861291c6c1fa9af410556eba7c8f7/href</a></iframe><img src="https://medium.com/_/stat?event=post.clientViewed&referrerSource=full_rss&postId=91871b2aa865" width="1" height="1" alt=""><hr><p><a href="https://medium.com/cloudzone/jenkinsx-deep-dive-meetup-91871b2aa865">JenkinsX Deep Dive (Meetup)</a> was originally published in <a href="https://medium.com/cloudzone">CloudZone</a> on Medium, where people are continuing the conversation by highlighting and responding to this story.</p>]]></content:encoded>
        </item>
        <item>
            <title><![CDATA[Running  AWS App Mesh with Elastic Kubernetes Service, CloudMap, XRay]]></title>
            <link>https://medium.com/cloudzone/running-aws-app-mesh-with-elastic-kubernetes-service-cloudmap-xray-19e25f4066d5?source=rss-c51152b94193------2</link>
            <guid isPermaLink="false">https://medium.com/p/19e25f4066d5</guid>
            <category><![CDATA[microservices]]></category>
            <category><![CDATA[aws-app-mesh]]></category>
            <category><![CDATA[service-mesh]]></category>
            <category><![CDATA[kubernetes]]></category>
            <category><![CDATA[observability]]></category>
            <dc:creator><![CDATA[Iftach Schonbaum]]></dc:creator>
            <pubDate>Thu, 09 Jul 2020 06:38:43 GMT</pubDate>
            <atom:updated>2020-07-09T06:49:11.029Z</atom:updated>
            <content:encoded><![CDATA[<h3>Running AWS App Mesh with Elastic Kubernetes Service, CloudMap, XRay</h3><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/0*6QD4haqpnrKRAgwG.png" /><figcaption>image from pullmanward3.org</figcaption></figure><p><em>Checkout this Webinar on running microservices in AWS App Mesh — the managed service mesh solution by AWS</em></p><p><em>The Session demonstrates the integrations with AWS EKS/ECS/Cloud Map/XRay — with live demos.</em></p><p><em>Comparison with Istio included.</em></p><p><em>Language: English</em></p><iframe src="https://cdn.embedly.com/widgets/media.html?src=https%3A%2F%2Fwww.youtube.com%2Fembed%2FXy7lclDIZj8%3Fstart%3D36%26feature%3Doembed%26start%3D36&amp;display_name=YouTube&amp;url=https%3A%2F%2Fwww.youtube.com%2Fwatch%3Fv%3DXy7lclDIZj8&amp;image=https%3A%2F%2Fi.ytimg.com%2Fvi%2FXy7lclDIZj8%2Fhqdefault.jpg&amp;key=a19fcc184b9711e1b4764040d3dc5c07&amp;type=text%2Fhtml&amp;schema=youtube" width="854" height="480" frameborder="0" scrolling="no"><a href="https://medium.com/media/22cf7686f1fc53167a933d2904a1b251/href">https://medium.com/media/22cf7686f1fc53167a933d2904a1b251/href</a></iframe><img src="https://medium.com/_/stat?event=post.clientViewed&referrerSource=full_rss&postId=19e25f4066d5" width="1" height="1" alt=""><hr><p><a href="https://medium.com/cloudzone/running-aws-app-mesh-with-elastic-kubernetes-service-cloudmap-xray-19e25f4066d5">Running  AWS App Mesh with Elastic Kubernetes Service, CloudMap, XRay</a> was originally published in <a href="https://medium.com/cloudzone">CloudZone</a> on Medium, where people are continuing the conversation by highlighting and responding to this story.</p>]]></content:encoded>
        </item>
        <item>
            <title><![CDATA[Apache Spark 3.0 Review–What the Spark is all about]]></title>
            <link>https://medium.com/cloudzone/apache-spark-3-0-review-what-the-spark-is-all-about-998844e12b3c?source=rss-c51152b94193------2</link>
            <guid isPermaLink="false">https://medium.com/p/998844e12b3c</guid>
            <category><![CDATA[sql]]></category>
            <category><![CDATA[analytics]]></category>
            <category><![CDATA[apache-spark]]></category>
            <category><![CDATA[big-data]]></category>
            <category><![CDATA[deep-learning]]></category>
            <dc:creator><![CDATA[Iftach Schonbaum]]></dc:creator>
            <pubDate>Thu, 21 Nov 2019 19:12:45 GMT</pubDate>
            <atom:updated>2019-11-26T15:26:05.930Z</atom:updated>
            <content:encoded><![CDATA[<h3>Apache Spark 3.0 Review — What the Spark is all about</h3><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*W_63I3zlKfi7hlIS12UD6g.png" /></figure><p>Apache Spark echo system is about to explode — Again! — this time with Sparks newest major version 3.0.</p><p>This article lists the new features and improvements to be introduced with <strong>Apache Spark 3.0</strong> — which its preview is <a href="https://spark.apache.org/news/spark-3.0.0-preview.html">already out</a> — very exciting!</p><p>Just for the stimulate — Alibaba Group competed with Spark 3.0 on the TPCDS benchmark and achieved the top spot!</p><p>Spark 3.0 is said to perform 17x faster compared with current versions on the TPCDS benchmark which is pretty impressive.</p><blockquote><strong>If you are a Spark user and you are familiar with Spark skip to the next section for the Spark 3.0 review</strong></blockquote><p>For those who are not yet familiar with Spark — it is a lightning fast, fault tolerant and in-memory general distributed data processing engine (kind of a MPP) that supports both batch and streaming processing and analytics.</p><p>Spark became very popular in late years and many organisations from small to enterprises adopted it including major cloud vendors where you can find a managed service that includes Spark.</p><p>Spark features APIs for distributed processing of structured and unstructured data like the RDD (Resilient Distributed Dataset), DataSet, DataFrame and more. it features SQL Capabilities including hooking up into any Hive metastore supporting catalogs (like AWS Glue) enabling great capabilities of explorations and analytics. On top of that it features Machine Learning capabilities (The Notorious Spark MLlib). Spark was meant to be a processing hub where it can connect many data sources — from relational and NoSQL databases, data lakes and warehouses and more — computing aggregations, data preprocessing and much more.</p><p>Spark 2.x introduced many improvements (Like Project Tungsten and the catalyst optimiser) and have made Spark shine out as a great tool and solution for ETL pipelines, Analytics in a Data Lake, engine for distributed machine learning training and serving, Streaming &amp; Structured Streaming (If mini-batches fits you) and more.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/479/1*2JB7AoY10zkkGmRRS8vSng.png" /><figcaption>Spark can hook into many sources supported by the DataSource API</figcaption></figure><p>Version 3.0 of spark is a major release and introduces major and important features:</p><h3>Language support</h3><p>Spark 3.0 will move to Python3 and Scala version is upgraded to version 2.12. In addition it will fully support JDK 11. Python 2.x is heavily deprecated .</p><h3>Adaptive execution of Spark SQL</h3><p>This feature helps in where statistics on the data source do not exists or are in accurate. So far Spark had some optimizations which could be set only in the planning phase and according to the statistics of data (e.g. the ones captured by the ANALYZE command when deciding weather to perform a Broadcast-hash join over an expensive Sort-merge join. In cases in which these statistics are missing or not accurate BHJ might not kick in. with adaptive execution in Spark 3.0 spark can examine that data at runtime once he had loaded it and opt-in to BHJ at runtime even it could not detect it on the planning phase.</p><h3>Dynamic Partition Pruning (DPP)</h3><p>Spark 3.0 introduces Dynamic Partition Pruning which is a major performance improvement for SQL analytics workloads that in term can make integration with BI tools much better. The idea behind DPP is to apply the filter set on the dimension table — mostly small and used in a broadcast hash join — directly on the fact table so it could skip scanning unneeded partitions.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/380/1*tWJBSFCjJa6LFNdqn4Ghew.png" /><figcaption>from Databricks session in Spark AI</figcaption></figure><p>DPP’s Optimisation is implemented both on the logical plan optimization and the physical planning. It showed speedup in many TCPDS queries and works well with star-schemas without the need to denormalize the tables.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/560/1*37ONcMVYYQusRz6s7S8eng.png" /><figcaption>Running TPCDS Query #98 with DPP. From Databricks session on DPP</figcaption></figure><p><a href="https://databricks.com/session_eu19/dynamic-partition-pruning-in-apache-spark">Databricks’s session on DPP.</a></p><h3>Enhanced Support for Deep Learning</h3><p>Deep Learning on Spark was already possible so far. However Spark MLlib was not focused on Deep Learning and did not offer deep learning algorithms and in particular didn&#39;t offer much for image processing. Existing projects like TensorFlowOnSpark, MMLSpark and some others made it possible somehow but presented significant challenges. For example — given that Spark resiliency is very good and knows to recompute tasks over partitions on failure — in deep learning for if you loose a partition in the middle of a training job and you recompute this individual partition Tensorflow or others will not work well. It requires to train on all partitions in the same time.</p><p>Spark 3.0 handles the above challenges much better. In addition it adds support for different GPUs like Nvidia, AMD, Intel and can use multiple types at the same time. In addition Vectorized UDFs can use GPUs for acceleration. For Kubernetes it offers GPU support in a flexible manner when running on Kubernetes.</p><h3>Better Kubernetes Integration</h3><p>Spark support for Kubernetes is relatively not matured in the 2.x version and difficult to use in production and performance was lacking in compare with the YARN cluster manager. Spark 3.0 introduces new shuffle service for Spark on Kubernetes that will allow dynamic scale up and down (more precisely out and in)</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/448/1*FbotyGuGFDAyQWpN0VP9sA.png" /></figure><p>Spark 3.0 also supports GPU support with pod level isolation for executors which makes scheduling more flexible on a cluster with GPUs. Spark authentication on Kubernetes also has some goodies.</p><h3>Graph features</h3><p>Graph processing can be used in data science for several application including recommendation engine and fraud detections.</p><p>Spark 3.0 introduces a whole new module named SparkGraph with major features for Graph processing. These features include the popular <a href="https://neo4j.com/developer/cypher-query-language/">Cypher query language</a> developed by <a href="https://neo4j.com/">Neo4J</a> which is a SQL like for graphs, the <a href="https://neo4j.com/developer/graph-database/">Property Graph Model</a> processed by this language and Graph algorithms. This integration is something Neo4J worked on for several years and it’s named Morpheus (formerly named Cypher for Spark) but as said will be named SparkGraph inside the spark components.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/298/1*SfrWLiXRZzqQlR3kjtMSRA.png" /></figure><p><a href="https://github.com/opencypher/morpheus">Morpheus</a> extends the Cypher language with multiple-graph feature, Graph catalog and Property graph data sources for integration with Neo4j engine, RDMS and more. It allows usage of the cypher language on graphs in a similar way SparkSQL operates over tabular data and it will have its own catalyst optimiser. In addition it would be possible to interoperate between SparkSQL and SparkGraph which can be very useful.</p><p>For a deep dive: <a href="https://www.youtube.com/watch?v=mEYvEF_2fDc">Check out this session</a></p><h3>ACID Transactions with Delta Lake</h3><p><a href="https://delta.io/">Delta Lake</a> is an open-source storage layer that brings ACID transactions to Apache Spark 3.0 and through easy implementation and upgrading of existing Spark applications, it brings reliability to Data Lakes. It announced to join the Linux foundation to grow its community.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/458/1*5yvA7_ODoj09TMibjPfgjw.png" /></figure><p>It solves issues presented when data in the data lake is modified simultaneously by multiple modifiers and allows you to focus on logic and not worry from inconsistencies. Its very valuable for streaming applications but also very relevant for batch scenarios. Over 3500 organizations already use Delta Lake.</p><p>Spark 3.0 supports data lake out of the box and can be used just as it is used for example with parquet. sometimes replacing the read class to the deltalake’s one is enough to start using Delta Lake.</p><p>Quick Start: <a href="https://docs.databricks.com/delta/quick-start.html">https://docs.databricks.com/delta/quick-start.html</a></p><h3>Growing integration with Apache Arrow data format</h3><p>Apache Arrow is an in-memory columnar data structure for efficient analytical operations. Its has benefits like being cross-language platform, performing zero-copy streaming messaging and interprocess communications without serialization costs which often occur with other systems.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/396/1*TUNvTYC20yshiwQtPECwJQ.png" /></figure><p>In Spark 3.0 Usage in Apache Arrow takes bigger place and its used to improve the interchange between the Java and Python VMs. This usage enables new features like Arrow accelerated UDFs, TensorFlow being able to read error data in CUDA and more features in the Deep Learning section in Spark 3.0</p><h3>Binary files data source</h3><p>Spark 3.0 supports binary file data source. You can use it like this:</p><pre>val df = spark.read.format(“binaryFile”)</pre><p>The above will read binary files and converts each one to a single row that contains the raw content and metadata of the file. The DataFrame will contain the following columns and possibly partition columns:</p><ul><li>path: StringType</li><li>modificationTime: TimestampType</li><li>length: LongType</li><li>content: BinaryType</li></ul><p>writing back a binary DataFrame/RDD is currently not supported.</p><h3>DataSource V2 Improvements</h3><p>Few improvements for the DataSource API are included with Spark 3.0:</p><ul><li>Pluggable catalog integration</li><li>Improved predicate push down for faster queries via reduced data loading</li></ul><p>In addition there are many JIRAs to solve many issues existing with the current DataSource API.</p><h3>YARN Features</h3><ul><li>Spark 3.0 can auto discover GPUs on a YARN cluster and schedule tasks specifically on nodes with GPUs.</li></ul><h3>More Features:</h3><p>The above features are somehow the major and more influencing one but Spark 3.0 ships more enhancements and features with it.</p><p>Mostly it is clear that Spark 3.0 is a big step up for data scientists and enables them to run Deep Learning with distributed training and serving.</p><h3>What’s being deprecated or removed</h3><p>More than a few things are being deprecated or removed. Make sure you read the deprecation and migration notes to see that you are covered the time you want to test your code with Spark 3.0.</p><h4>Python 2.x</h4><p>Python 2.7 will still work, but not anymore tested in the release lifecycle of Spark. this effectively means — you DON’T use it. This is relevant for all Panadas and NumPy users in Python 2. So from Spark 3.0 PySpark users will see a deprecation warning if Python 2 is used. A migration guide for PySpark users to migrate to Python 3 will be available. Completely removing Python 2 support is expected to occur in 1.1.2020 when PySpark uses will see an error in case Python 2 is used. After this date patches for Python 2 might be rejected for Spark versions 2.4.x</p><h4>Spark MLlib</h4><p>I coudnt totaly figure what will be the future of SparkMLlib with Apache Spark 3.x versions. Ill update this blog asap once I know more.</p><h4>&lt;I’ll Add more deprecations/removals here when I know&gt;</h4><p>I hope you liked this thorough review and that it was helpful. Please feel free to leave any questions on the comment sections.</p><p>I’de like to thank:</p><ul><li><a href="https://medium.com/u/5e502786e0a2">holden karau</a> for continuously being a source of inspiration and knowledge. Parts of this article comes from his talks and posts on the subject.</li><li><a href="https://medium.com/u/5ae67e7eecef">Databricks</a> for giving great sessions and leading innovations around Apache Spark — also being a source for this post.</li></ul><img src="https://medium.com/_/stat?event=post.clientViewed&referrerSource=full_rss&postId=998844e12b3c" width="1" height="1" alt=""><hr><p><a href="https://medium.com/cloudzone/apache-spark-3-0-review-what-the-spark-is-all-about-998844e12b3c">Apache Spark 3.0 Review–What the Spark is all about</a> was originally published in <a href="https://medium.com/cloudzone">CloudZone</a> on Medium, where people are continuing the conversation by highlighting and responding to this story.</p>]]></content:encoded>
        </item>
        <item>
            <title><![CDATA[CNCF: What’s Incubating — Part 1: NATS Messaging]]></title>
            <link>https://medium.com/cloudzone/cncf-whats-incubating-part-1-nats-messaging-ac5c0af317e0?source=rss-c51152b94193------2</link>
            <guid isPermaLink="false">https://medium.com/p/ac5c0af317e0</guid>
            <category><![CDATA[messaging]]></category>
            <category><![CDATA[cloud-native]]></category>
            <category><![CDATA[cloud-computing]]></category>
            <category><![CDATA[kubernetes]]></category>
            <category><![CDATA[cncf]]></category>
            <dc:creator><![CDATA[Iftach Schonbaum]]></dc:creator>
            <pubDate>Wed, 23 Oct 2019 15:07:57 GMT</pubDate>
            <atom:updated>2019-10-23T17:08:11.603Z</atom:updated>
            <content:encoded><![CDATA[<h3>CNCF: What’s Incubating — Part 1: NATS Messaging</h3><figure><img alt="" src="https://cdn-images-1.medium.com/max/760/0*MtgNdqivL20fX2RQ.jpg" /></figure><p>The Cloud Native Computing Foundation is the home for Cloud Native projects which seats right under the Linux foundation.</p><p>It hosts very well known projects (and for some their CI/CD) like Kubernetes, Prometheus, Envoy, Fluentd and more. Well, these are the graduated ones. Graduated projects are projects that are passed their major adoption levels and been voted for and considered very stable.</p><p>Project creators and communities has high incentives to host their project under the CNCF.</p><blockquote>“In a world where GitHub use is ubiquitous, it is no longer sufficient for a software foundation to offer just a software repo, mailing list, and website. An enhanced set of services is required as it facilitates increased adoption.”</blockquote><p>CNCF provides the projects hosted by it with many services and incentives as money investments, community tooling and foundations, program management, event management, marketing services, certifications, legal services and more.</p><p>As part of our work at CloudZone, a Premier consulting partner of all the major cloud providers and CNCF Silver Member and a CNCF Kubernetes Certified Service Provider, we include many of the CNCF tools and frameworks in our solutions.</p><p>This series of blog posts is intended to summarize in detail the CNCF hosted projects that are currently in the incubation stage (after sandboxing, before graduating).</p><h3>NATS Messaging</h3><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/0*oiot7kRx44RMymWZ.png" /></figure><p>NATS is a lightweight cloud native messaging system for next generation cloud native distributed applications, edge computing platforms and devices. It is already an <strong>10-years</strong> old and is production proven (Originally Built to power CloudFoundry)</p><p>NATS is:</p><ul><li>Highly Resilient</li><li>Highly Secured</li><li>Highly scalable with built in load balancing and auto scaling</li><li>Extremely lightweight</li><li>frictionless for developers in an agile environment</li><li>Provides QoS for messaging</li><li>Support for 30 client SDKs</li></ul><p>NATS is focused on performance, security, simplicity , and availability. Its community has grown in the last 2 years.</p><h4>Use Cases:</h4><ul><li><strong>Cloud Messaging:</strong> Microservices Transport, Control Plane, Service Discovery and Event sourcing</li><li><strong>IoT and Edge</strong></li><li><strong>Mobile</strong></li><li><strong>High Fan-out messaging</strong></li><li><strong>Augmenting or replacing legacy messaging systems</strong></li></ul><blockquote><strong>Core Entity: Subject: </strong>You can think of it as a Topic to which any client can subscribe. It is a string representing an interest in data and subscribers can subscribe to subjects using wildcards that matches these subject strings.</blockquote><h4>Messaging Patterns</h4><p><strong><em>Pub/Sub: </em></strong>publish a message to a subject. Subscribers on this subject will receive this message. Used for high fanout and parallelization of work.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/974/1*OzuNaAps78j0llaUHiVVoQ.png" /><figcaption><a href="https://nats-io.github.io/">nats.io</a></figcaption></figure><p><strong><em>Load balanced queue subscribers: </em></strong>When you create subscribers you can add them to a group. When a message is published, one of them will receive it. You can think of it as a Kafka Consumer Group. Used for load balancing, auto-scaling, lame duck mode during upgrades.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*xb3Ts_zVe6G53ToUvs2rdg.png" /><figcaption><a href="https://nats-io.github.io/">https://nats-io.github.io</a></figcaption></figure><p><strong><em>Request/Reply: </em></strong>Unique subjects that enables request/reply patterns where you can send request to many subscriber and receive only 1 response from the NATS cluster thus getting the fastest response with least latency. In this case NATS will prevent the messaging from continuously propagating after a response is been sent. Used for request to many and handling first response where you can scale the subscribers to achieve faster response.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/974/1*o53uqojUDkjqFtxnsXohdg.png" /><figcaption><a href="https://nats-io.github.io/">https://nats-io.github.io</a></figcaption></figure><h4>Graceful shutdown using Drain API</h4><p>Clients can invoke graceful shutdown using the drain API where the client unsubscribes and stop receiving new messages but continue to process buffered ones. Used to prevent data loss in shutting down or scale-down events and in application upgrades.</p><h4>Availability &amp; Resiliency</h4><p>NATS prioritises the health of the system as a whole over individual client or server. For example, when a NATS client is not consuming fast enough the NATS server will cut this client off, considering it as not healthy. The NATS cluster is a full mesh of NATS server where any of them can go down and the rest will take over its clients. Also, connections of servers and clients are self healing meaning they will try to reconnect. NATS was proved to be very stable and running under load for long duration of times without any interruption.</p><h4>Performance</h4><p>Extremely fast: 18 million messages per second with single server and single data stream and up to 80 million per second with multiple streams. It is very scalable.</p><h4>Simplicity</h4><p>NATS is single binary proved with 8MB docker image. the text-protocol payloads are binary. Configuration is not more than a url and credentials. Servers are auto-discoverable and share discovered topology and configuration is shared — which is all transparent to the clients. The client APIs are simple and straight forward.</p><p>NATS has a prometheus exporter for exposing metrics and a Fluentd plugin.</p><h4>Delivery modes</h4><p><strong>NATS support two delivery modes:</strong></p><ul><li>At most once (Core) where there is no guarantee of message delivery. in this case application must detect and handle lost messages</li><li>At least once (Streaming) where a message will always be delivered and in certain cases more than once.</li></ul><p>NATS by choice does not provides Exactly once mode as it is considered unnecessary and is slow and complex which does not stands together with one of the core missions of NATS where it focuses on simplicity and high performance.</p><p>Streaming also includes features as replaying by time or sequence number, rate matching per subscriber, storage tiers (memory, file ,database), HA and scale through partitioning.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/770/0*cn2YJQoP_iPrjsq5.png" /></figure><h4>Multi-tenancy</h4><p>Multi-tenancy is possible in NATS via accounts and enriched with Services and Streams.</p><blockquote><strong>Accounts — </strong>Isolated communication contexts with which you can use a single NATS deployment for multiple isolated operators. Accounts can share data with other accounts via Services and Streams.</blockquote><blockquote><strong>Service — </strong>RPC endpoints that enables a Request/Reply delivery pattern between accounts — one on one conversation. Used for: monitoring probes, certificate generation services, secure vault.</blockquote><blockquote><strong>Streams — </strong>Enables publish/subscribe pattern between Accounts. Used for: Global Alerts, Slack like solutions, Twitter feeds.</blockquote><p>For example, given two clients that publish and subscribe on Subject X in account A , no Subscriber of Subject X in account B will receive this message unless there is a Stream allowing it. The great things is that for all of the above no client configuration is needed.</p><h4>Security</h4><p>NATS is secured with:</p><ul><li>Authentication via TLS certificates,, basic credentials &amp; NKeys (based on ED25519) and JWTs</li><li>Encryption with TLS</li><li>Policy</li><li>Allow/Deny based Subject authorization with wildcard support</li></ul><p>All updated to these entities are with zero down-time.</p><h4>NATS Global Deployment with SuperClusters</h4><p>Clusters of NATS clusters can be used for global implementation of NATS messaging. In its core you can use a global distributed load balanced queue subscribers which is geo aware. For example, given a queued subscribers in Europe and in the US, if i publish a message from Europe, NATS will deliver the message to one of the subscribers in Europe.</p><h4>More goodies:</h4><ul><li>Integration with more messaging projects like Kafka</li><li>Native MQTT Support &amp; Microcontroller clients for IoT</li><li>WebSocket Support</li></ul><h4>Integration with Kubernetes</h4><p>Integration via the NATS Operator which creates and manages NATS cluster. The Operator uses K8s RBAC for authorization with service accounts. Provides Hot reload of Secrets stored configuration.</p><p>An Helm chart is available for installing on Kubernetes</p><h4><strong>Among NATS Adopters</strong></h4><figure><img alt="" src="https://cdn-images-1.medium.com/max/580/1*EI9ptLj7UIwWqVNi0a1mdQ.png" /></figure><img src="https://medium.com/_/stat?event=post.clientViewed&referrerSource=full_rss&postId=ac5c0af317e0" width="1" height="1" alt=""><hr><p><a href="https://medium.com/cloudzone/cncf-whats-incubating-part-1-nats-messaging-ac5c0af317e0">CNCF: What’s Incubating — Part 1: NATS Messaging</a> was originally published in <a href="https://medium.com/cloudzone">CloudZone</a> on Medium, where people are continuing the conversation by highlighting and responding to this story.</p>]]></content:encoded>
        </item>
        <item>
            <title><![CDATA[XGBoost Distributed Training and Parallel Predictions with Apache Spark]]></title>
            <link>https://medium.com/cloudzone/xgboost-distributed-training-and-predicting-with-apache-spark-1127cdfb31ae?source=rss-c51152b94193------2</link>
            <guid isPermaLink="false">https://medium.com/p/1127cdfb31ae</guid>
            <category><![CDATA[xgboost]]></category>
            <category><![CDATA[machine-learning]]></category>
            <category><![CDATA[apache-spark]]></category>
            <category><![CDATA[spark]]></category>
            <category><![CDATA[big-data]]></category>
            <dc:creator><![CDATA[Iftach Schonbaum]]></dc:creator>
            <pubDate>Sat, 05 Oct 2019 20:32:37 GMT</pubDate>
            <atom:updated>2019-10-07T12:42:36.775Z</atom:updated>
            <content:encoded><![CDATA[<figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/0*GuTWDCReVWYFdZYh.png" /></figure><h4>Background</h4><p>In Boosting (ML ensemble method), algorithms implement a sequential process (as opposed to Bagging where it is parallelised) that generates weak learners and combine them to a strong learner (as in all Ensemble methods). In Boosting, at each iteration of this process, the model tries to correct the mistakes of the previous one in adaptive way — Unlike Bagging in which weak learners are trained independently.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/515/0*O-kWynolNcvXCWjx.png" /></figure><p>One of the Boosting algorithms, Gradient Boosting, is using the gradient descent to minimise the loss function directly in these sequential models (As opposed to another boosting algorithm, AdaBoosting, where fitting is achieved by changing weights of training instances)</p><p>The weak learners created in Gradient boosting as part of training are usually implemented with decision trees. The major inefficiency in Gradient Boosting is that the process of creating these trees is <strong>sequential</strong> — i.e. it creates one decision tree at a time.</p><p>To overcome that, an extension of Gradient Boosting was introduced (by Tianqi Chen and Carlos Guestrin) named XGBoost which stands for Extreme Gradient Boosting. It’s kind of Gradient Boosting on steroids and is used for mainly classification, but also regression and ranking.</p><p>It introduces lots of performance enhancements through hyper-parameters, GPU support, cross validation capabilities &amp; algorithm regularisations over the tradition Gradient Boosting to make the overall model more efficient, faster to train and less prune to overfitting.</p><p>XGboost became popular in the last years and and won many of the machine learning competitions at Kaggle and considered to have more computational power and accuracy.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/777/0*bpEDPjgIvbJOPaWz.png" /></figure><h4>XGBoost with Apache Spark</h4><p>A common workflow in ML is to utilize systems like Spark to construct ML Pipeline in which you preprocess and clean data, and pass the results to the machine learning phase, usually with Spark MLlib once you already use Spark.</p><p>In the context of this article the important feature XGBoost introduces is parallelism for the tree building — it essentially enables distributed training and predicting across nodes. This means that if I am an Apache Spark MLlib user or company, i could as well use it to empower XGBoost training and serving in a production grade way and enjoy both the high performant algorithm of XGboost and Spark’s powerfull processing engine for feature engineering and constructing ML pipelines.</p><p>Meet <strong>XGBoost4J-Spark — </strong>a project that integrates XGBoost and Apache Spark by fitting XGBoost to Apache Spark’s MLlIB framework.</p><p>XGBoost4J-Spark makes it possible to construct a MLlib pipeline that preprocess data to fit for XGBoost model, train it and serve it in a distributed fashion for predictions in production. With this library each XGBoost worker is wrapped by a Spark task and the training dataset in Spark’s memory space is sent to XGBoost workers that live inside the spark executors in a transparent way.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1013/1*2TwpcfdhkHqoDmi71qEY7g.png" /></figure><p>To write a ML XGBoost4J-Spark application you first need to include its dependency:</p><pre>&lt;dependency&gt;<br>    &lt;groupId&gt;ml.dmlc&lt;/groupId&gt;<br>    &lt;artifactId&gt;xgboost4j-spark&lt;/artifactId&gt;<br>    &lt;version&gt;0.90&lt;/version&gt;<br>&lt;/dependency&gt;</pre><h4>Data Preparation (Iris example)</h4><p>As said, XGBoost4J-Spark enables fitting of data to the XGBoost interface.</p><p>Once we have the Iris dataset read into a DataFrame we need to:</p><ol><li>Transform String-typed label cols to Double-typed label.</li><li>Assemble the feature columns as a vector to fit to the data interface of Spark ML framework.</li></ol><pre><strong>import</strong> <strong>org.apache.spark.ml.feature.StringIndexer<br>import</strong> <strong>org.apache.spark.ml.feature.VectorAssembler</strong></pre><pre><strong>val</strong> stringIndexer <strong>=</strong> <strong>new</strong> <strong>StringIndexer().</strong><br>  setInputCol<strong>(</strong>&quot;class&quot;<strong>).</strong><br>  setOutputCol<strong>(</strong>&quot;classIndex&quot;<strong>).</strong><br>  fit<strong>(irisDF)</strong></pre><pre><strong>val</strong> labelTransformed <strong>=</strong> stringIndexer<strong>.</strong>transform<strong>(irisDF).</strong>drop<strong>(</strong>&quot;class&quot;<strong>)<br>val</strong> vectorAssembler <strong>=</strong> <strong>new</strong> <strong>VectorAssembler().</strong><br>  setInputCols<strong>(Array(</strong>&quot;sepal length&quot;<strong>,</strong> &quot;sepal width&quot;<strong>,</strong> &quot;petal length&quot;<strong>,</strong> &quot;petal width&quot;<strong>)).</strong><br>  setOutputCol<strong>(</strong>&quot;features&quot;<strong>)</strong></pre><pre><strong>val</strong> xgbInput <strong>=</strong> vectorAssembler<strong>.</strong>transform<strong>(</strong>labelTransformed<strong>).</strong>select<strong>(</strong>&quot;features&quot;<strong>,</strong> &quot;classIndex&quot;<strong>)<br></strong></pre><p>The above results in a DataFrame with only two columns, “features”: vector-representing the iris features and “classIndex”: Double-typed label. A DataFrame like this can be fed to XGBoost4J-Spark’s training engine directly.</p><h4>Distributed Training</h4><pre><strong>import</strong> <strong>ml.dmlc.xgboost4j.scala.spark.XGBoostClassifier</strong><br><br><strong>val</strong> xgbClassifier <strong>=</strong> <strong>new</strong> <strong>XGBoostClassifier().</strong><br>      setFeaturesCol<strong>(</strong>&quot;features&quot;<strong>).<br>      </strong>setLabelCol<strong>(</strong>&quot;classIndex&quot;<strong>).<br>      </strong>setObjective(&quot;multi:softmax&quot;<strong>)<br></strong>      setMaxDepth<strong>(2).<br>      </strong>setNumClass<strong>(3).<br>      </strong>setNumRound<strong>(100).<br>      </strong>setNumWorkers<strong>(</strong>10<strong>).</strong></pre><p>for a full list of XGBoost parameters see <a href="https://xgboost.readthedocs.io/en/latest/parameter.html">here</a>. Note that in XGBoost4J-Spark you can use also the camel case format as seen above.</p><p><strong>Notes:</strong></p><ol><li><strong>multi:softmax</strong> objective means we are doing multiclass classification using softmax. This requires to set the number of classes using the <strong>num_class</strong> param.</li></ol><p>2. <strong>max_depth </strong>is the maximum depth of a tree created in each boosting iteration. Increasing this value will make the model more complex and more likely to overfit. XGBoost consumes lots of memory when training deep trees.</p><p>3. <strong>num_rounds </strong>is the number of rounds for boosting.</p><p>4. The <strong>num_workers</strong> parameter controls how many parallel workers we want to have when training a XGBoostClassificationModel. This will later on translated to Spark pending tasks which in term will be handled by the cluster manager (YARN in most cases).</p><p>Early Stopping is supported using the <strong>num_early_stopping_rounds</strong> and <strong>maximize_evaluation_metrics</strong> parameters.</p><p>Now we can create the transformer by fitting XGBoost Classifier with the input DataFrame. This is essentially the training process that yields the model that can be used in prediction.</p><pre><strong>val</strong> xgbClassificationModel <strong>=</strong> xgbClassifier<strong>.</strong>fit<strong>(</strong>xgbInput<strong>)</strong></pre><h4>Parallel Prediction</h4><p>XGBoost4j-Spark supports batch prediction and single instance prediction.</p><p>For batch prediction, the model takes a DataFrame with a column containing feature vectors, predict for each feature vector, and output a new DataFrame with the results. In this process XGBoost4J-Spark starts a Spark task containing a XGBoost worker for each partition of the input DataFrame for parallel prediction of the batch.</p><pre><strong>val</strong> <strong>predictionsDf</strong> <strong>=</strong> xgbClassificationModel<strong>.</strong>transform<strong>(inputDF)<br>predictionsDf.show()</strong></pre><pre>+----------------+----------+-------------+-------------+----------+<br>|       features |classIndex|rawPrediction| probability |prediction|<br>+----------------+----------+-------------+-------------+----------+<br>|[5.1,3.5,1.2,.. |       0.0|[3.4556984...|[0.9957963...|       0.0|<br>|[4.7,3.2,1.3,.. |       0.0|[3.4556984...|[0.9961891...|       0.0|<br>|[5.7,4.4,1.5,.. |       0.0|[3.4556984...|[0.9964334...|       0.0|<br>+----------------+----------+-------------+-------------+----------+</pre><p>For single prediction, the model accepts a single Vector.</p><pre><strong>val</strong> features <strong>=</strong> xgbInput<strong>.</strong>head<strong>().</strong>getAs<strong>[Vector](</strong>&quot;features&quot;<strong>)</strong><br><strong>val</strong> result <strong>=</strong> xgbClassificationModel<strong>.</strong>predict<strong>(</strong>features<strong>)</strong></pre><p>Single prediction with XGBoost is not recommended because of the overhead that will be triggered internally compared with the just one prediction.</p><p>The latest release (0.9) of XGBoost’s XGBoost4J-Spark now requires Spark 2.4.x. mainly because it uses facilities of org.apache.spark.ml.param.shared which are not fully available on earlier versions of Spark.</p><p>This version also includes more consistent handling of missing values, better performance for multi-core CPUs, better control for caching partitioned training data to reduce training time and more.</p><p>for more information about XGBoost check out the <a href="https://xgboost.readthedocs.io/en/latest/index.html">docs</a>.</p><h4>References:</h4><ol><li>XGBoost with CUDA</li></ol><p><a href="https://devblogs.nvidia.com/gradient-boosting-decision-trees-xgboost-cuda/">Gradient Boosting, Decision Trees and XGBoost with CUDA | NVIDIA Developer Blog</a></p><p>2. XGBoost in spark with GPU with RAPIDS XGboost4J-Spark</p><p><a href="https://github.com/rapidsai/spark-examples">rapidsai/spark-examples</a></p><img src="https://medium.com/_/stat?event=post.clientViewed&referrerSource=full_rss&postId=1127cdfb31ae" width="1" height="1" alt=""><hr><p><a href="https://medium.com/cloudzone/xgboost-distributed-training-and-predicting-with-apache-spark-1127cdfb31ae">XGBoost Distributed Training and Parallel Predictions with Apache Spark</a> was originally published in <a href="https://medium.com/cloudzone">CloudZone</a> on Medium, where people are continuing the conversation by highlighting and responding to this story.</p>]]></content:encoded>
        </item>
    </channel>
</rss>