<?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 Das Binodeswar on Medium]]></title>
        <description><![CDATA[Stories by Das Binodeswar on Medium]]></description>
        <link>https://medium.com/@das.binodeswar217?source=rss-d7ba0954c12d------2</link>
        <image>
            <url>https://cdn-images-1.medium.com/fit/c/150/150/0*5CYMR0G8ESKtI7S_</url>
            <title>Stories by Das Binodeswar on Medium</title>
            <link>https://medium.com/@das.binodeswar217?source=rss-d7ba0954c12d------2</link>
        </image>
        <generator>Medium</generator>
        <lastBuildDate>Wed, 18 Nov 2026 00:01:55 GMT</lastBuildDate>
        <atom:link href="https://medium.com/@das.binodeswar217/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[User-Specific Rate Limiting in System Design]]></title>
            <link>https://medium.com/@das.binodeswar217/user-specific-rate-limiting-in-system-design-c7d5903a11e8?source=rss-d7ba0954c12d------2</link>
            <guid isPermaLink="false">https://medium.com/p/c7d5903a11e8</guid>
            <category><![CDATA[website-traffic]]></category>
            <category><![CDATA[rate-limiter]]></category>
            <category><![CDATA[rate-limit]]></category>
            <category><![CDATA[software-development]]></category>
            <dc:creator><![CDATA[Das Binodeswar]]></dc:creator>
            <pubDate>Thu, 28 May 2026 02:02:58 GMT</pubDate>
            <atom:updated>2026-05-28T02:02:58.644Z</atom:updated>
            <content:encoded><![CDATA[<p>User-Specific Rate Limiting in System Design</p><p>Most of us first learn rate limiting as:</p><p>“Allow X requests per second per IP.”</p><p>Simple. Easy. Effective.</p><p>Until real production traffic arrives.</p><p>IP-based rate limiting starts breaking in modern systems because:</p><p>* thousands of users may sit behind the same NAT/proxy<br>* mobile networks constantly rotate IPs<br>* attackers can bypass limits using distributed traffic</p><p>That’s why modern systems increasingly move toward &#39;user-specific rate limiting&#39; — enforcing limits based on authenticated identity (`user_id`, API key, token) instead of network address.</p><p>And once you build it at scale, you realize rate limiting is far more than just counting requests.</p><p>A few things that become critical very quickly:</p><p>Shared counters across instances<br>Local in-memory counters fail immediately in distributed systems. This is why Redis becomes the default choice.</p><p>Atomic operations<br>Increment + expiry must be atomic, otherwise concurrent requests create inconsistent enforcement.</p><p>Choosing the right algorithm</p><p>* Fixed Window → simple but burst-prone<br>* Sliding Window → accurate but expensive<br>* Token Bucket → practical and burst-friendly</p><p>Most production systems eventually lean toward Sliding Window Counter or Token Bucket.</p><p>Multi-region tradeoffs<br>Strict global rate limiting sounds good architecturally… until cross-region latency becomes the bottleneck.</p><p>Many large systems prefer approximate enforcement with regional quotas because availability and latency matter more than perfect precision.</p><p>One interesting realization while designing these systems:</p><p>A good rate limiter is also an observability and business signal.</p><p>Users constantly hitting limits may indicate:</p><p>* abusive traffic<br>* inefficient integrations<br>* power users<br>* upsell opportunities</p><p>So rate limiting stops being just a protection layer and becomes part of product intelligence.</p><p>The deeper you go into distributed systems, the more you realize:<br>simple-looking infrastructure components usually hide the hardest engineering tradeoffs underneath.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*e6gvvcFwD65ExD9VaUKXvw.png" /></figure><p>#SystemDesign #DistributedSystems #BackendEngineering #Scalability #Redis #API #SoftwareEngineering #CloudComputing</p><img src="https://medium.com/_/stat?event=post.clientViewed&referrerSource=full_rss&postId=c7d5903a11e8" width="1" height="1" alt="">]]></content:encoded>
        </item>
        <item>
            <title><![CDATA[DuckDB: The Simpler Way to Do Analytics]]></title>
            <link>https://medium.com/@das.binodeswar217/duckdb-the-simpler-way-to-do-analytics-d342dd7930e2?source=rss-d7ba0954c12d------2</link>
            <guid isPermaLink="false">https://medium.com/p/d342dd7930e2</guid>
            <category><![CDATA[analytics]]></category>
            <category><![CDATA[duckdb]]></category>
            <category><![CDATA[software-development]]></category>
            <dc:creator><![CDATA[Das Binodeswar]]></dc:creator>
            <pubDate>Sat, 25 Apr 2026 15:56:04 GMT</pubDate>
            <atom:updated>2026-04-26T04:49:58.371Z</atom:updated>
            <content:encoded><![CDATA[<p><strong>DuckDB: The Simpler Way to Do Analytics</strong></p><p>We’ve made data analytics unnecessarily heavy.</p><p>Clusters, warehouses, pipelines, orchestration—often before we even understand the problem. But not every use case needs that level of complexity.</p><p>Sometimes, you just want to query your data and move on.</p><p>That’s exactly where DuckDB fits in.</p><p><strong>What it actually is</strong></p><p>DuckDB is an in-process analytical database built for fast OLAP workloads.</p><p>A simple way to think about it:</p><p>Like SQLite, but for analytics</p><p>Runs inside your app or notebook</p><p>No server, no setup, no overhead</p><p>You don’t deploy DuckDB. You just use it.</p><p><strong>Why it feels different</strong></p><p>Most data tools require you to move data before you can analyze it.</p><p>DuckDB flips that.</p><p>You can query data directly from files:</p><p>SELECT * FROM &#39;data.parquet&#39; WHERE revenue &gt; 1000;</p><p>No ingestion. No ETL. No waiting.</p><p>It works directly on Parquet, CSV, JSON—treating files as queryable data sources.</p><p><strong>What makes it quietly powerful</strong></p><p>Fast by design<br>Columnar execution and vectorized processing make analytical queries fast—even on large datasets.</p><p>No infrastructure to think about<br>No servers. No clusters. It runs wherever your code runs.</p><p>Plays well with your tools<br>Works seamlessly with Pandas, NumPy, and Apache Arrow.<br>You can run SQL directly on DataFrames without copying data.</p><p>Built for modern data formats<br>Parquet isn’t an afterthought—it’s central to how DuckDB operates.</p><p>Where it naturally fits</p><p>DuckDB isn’t trying to replace your warehouse. It removes steps before you even need one.</p><p>It works best for:</p><p>Exploratory analysis</p><p>Local data processing</p><p>Data science workflows</p><p>Lightweight ETL</p><p>If you’re iterating quickly, it just feels right.</p><p>Where you should think twice</p><p>It’s not built for:</p><p>High-concurrency systems</p><p>Real-time transactions</p><p>Multi-user production databases</p><p>For those, tools like PostgreSQL or cloud warehouses are still the better choice.</p><p>A shift in mindset</p><p>DuckDB represents a simple but powerful idea:</p><p>Don’t move data unless you have to.</p><p>Instead of shipping data to a warehouse, you bring compute closer to where the data already lives.</p><p>Less movement. Less cost. Less complexity.</p><p>A practical perspective</p><p>DuckDB won’t replace your entire stack—and it shouldn’t.</p><p>But it will quietly reduce how much infrastructure you actually need.</p><p>You start with quick analysis. Then prototyping. Then small pipelines.</p><p>And somewhere along the way, you realize you’re solving problems faster—with fewer moving parts.</p><p>That’s not just convenience. That’s leverage.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*TbxydjBYnD63Q7EkmobDag.png" /></figure><img src="https://medium.com/_/stat?event=post.clientViewed&referrerSource=full_rss&postId=d342dd7930e2" width="1" height="1" alt="">]]></content:encoded>
        </item>
        <item>
            <title><![CDATA[Software Architecture Is a Spectrum, Not a Switch]]></title>
            <link>https://medium.com/@das.binodeswar217/software-architecture-is-a-spectrum-not-a-switch-ae6c019a819b?source=rss-d7ba0954c12d------2</link>
            <guid isPermaLink="false">https://medium.com/p/ae6c019a819b</guid>
            <dc:creator><![CDATA[Das Binodeswar]]></dc:creator>
            <pubDate>Tue, 24 Mar 2026 17:41:57 GMT</pubDate>
            <atom:updated>2026-03-24T17:41:57.854Z</atom:updated>
            <content:encoded><![CDATA[<p>Software Architecture Is a Spectrum, Not a Switch</p><p>Over the years, I’ve noticed a pattern in how we talk about software architecture. We tend to frame everything as a binary choice:</p><p>Monolith vs Microservices</p><p>Stateless vs Stateful</p><p>REST vs RPC</p><p>But when you actually build and run systems in production, you realize pretty quickly… nothing is that absolute.</p><p>The Reality Behind the Theory</p><p>On paper, things look clean:</p><p>Stateless systems scale better</p><p>RESTful APIs are the standard</p><p>Microservices give flexibility</p><p>In real life, though:</p><p>“Stateless” systems still depend on databases, caches, tokens</p><p>“REST” APIs often bend rules to meet business needs</p><p>“Microservices” can become tightly coupled if not designed carefully</p><p>So what are we really doing? We’re constantly making trade-offs, not following purity. Architecture isn’t a one-time decision. It evolves. Think of it more like a journey:</p><p>Monolith → Modular Monolith → Distributed Monolith → Microservices</p><p>Most systems don’t jump to microservices overnight—and honestly, they shouldn’t.</p><p>The Trap I See Most Teams Fall Into</p><p>The “distributed monolith.” It usually starts with good intentions:</p><p>Break the app into services</p><p>Scale independently</p><p>Move faster</p><p>But ends up like this:</p><p>Shared databases</p><p>Tight coupling</p><p>Too many synchronous calls</p><p>Now you have:</p><p>Higher infra cost</p><p>More operational overhead</p><p>Same level of dependency</p><p>Basically, the complexity of microservices without the benefits.</p><p>What Has Worked Better in Practice</p><p>A pattern I’ve seen work well: Start with a modular monolith. Extract only what truly needs to be a service. Typical examples:</p><p>Authentication</p><p>Payments</p><p>High-traffic or high-risk modules</p><p>Everything else? Keep it together. Keep it simple.</p><p>A Better Question to Ask</p><p>Instead of asking:</p><p>“Should we move to microservices?”</p><p>Try asking:</p><p>“Where do we actually need boundaries?”</p><p>Only introduce services when you genuinely need:</p><p>Independent scaling</p><p>Clear ownership</p><p>Faster, isolated deployments</p><p>Different SLAs</p><p>A Simple Approach That Scales</p><p>Build a modular monolith</p><p>Keep boundaries clean inside the codebase</p><p>Watch where the system feels pain</p><p>Extract services when the need is real</p><p>Good architecture isn’t about picking a side. It’s about making decisions that fit our context today—and can evolve tomorrow.</p><p>Because in software… the best solutions are rarely black or white. They live in the grey.</p><img src="https://medium.com/_/stat?event=post.clientViewed&referrerSource=full_rss&postId=ae6c019a819b" width="1" height="1" alt="">]]></content:encoded>
        </item>
    </channel>
</rss>