Shalin Shah

Have you ever noticed those WhatsApp moments where replies pop up before the messages? It's like getting the punchline before the joke! But seriously, forgetting about time sync could turn software into a comedy of errors in a distributed system. Time sync might sound boring, but it's the unsung hero of the distributed world! Time synchronization is very critical in a distributed environment; below are some of the Important time synchronization concepts in distributed systems to know. Certainly! Here are brief descriptions for each topic suitable for a LinkedIn post: 1. Lamport Clock - The Lamport Clock is a logical clock used in distributed systems to order events based on occurrence. It assigns timestamps to events, enabling a partial ordering of events in distributed computing. 2. Google True Time - Google True Time is a highly accurate timekeeping service developed by Google. It utilizes a combination of atomic clocks and GPS receivers to offer precise time information for distributed systems. 3. Vector Clock - Vector Clock is a mechanism used for tracking causality in distributed systems. It helps determine the relative ordering of events across different nodes by assigning vector timestamps. 4. Marzullo's Algorithm - Marzullo's Algorithm is utilized to find the common intersection of time intervals in a distributed system. It efficiently identifies the period when multiple intervals overlap, aiding event synchronization. 5. Network Time Protocol (NTP) - NTP is a network protocol for clock synchronization between computer systems over packet-switched, variable-latency data networks. It maintains time synchronization accuracy within a few milliseconds. 6. Precision Time Protocol (PTP) - PTP is a protocol for high-precision time synchronization in distributed systems, aiming for sub-microsecond or nanosecond-level accuracy. It ensures highly accurate clock synchronization among nodes 7. Cristian's Algorithm - Cristian's Algorithm is an early approach to synchronizing clocks in a distributed system. It involves a central time server providing time to clients and helps with overall synchronization 8. Hybrid Clock A hybrid Clock combines system & logical time to create a more accurate and reliable timekeeping system. It leverages the strengths of various clock mechanisms for improved accuracy. 9. Berkeley's Algorithm - Berkeley's Algorithm is used to synchronize clocks in distributed systems. It involves a central coordinator adjusting clocks based on an average time, reducing time differences among machines. 10. GPS & Atomic Clock - GPS satellites use atomic clocks to provide highly accurate time signals for synchronization. Combining GPS with atomic clocks offers precise timing information crucial for various applications. #systemdesign #engineering #productdevelopment #distributedsystems #microservices

26

7 Comments

Not only do you get one click deploy of gpt-oss on GKE but also you get detailed benchmarks to make informed decisions on how to deploy. Customers can save months on qualification. Best part it's based on an OSS benchmarking framework so you get standard and repeatable deployments. https://lnkd.in/geu_NU5U cc Sean Horgan Ashok Chandrasekar Nathan Beach Iftach Ragoler Drew Bradstock Connor McCoy

43

3 Comments

I wrote five markdown files on product principles this week and put them in every Aurigo repo. Kevin Koenig our CTO, Shashi Kumar Balu our VP Product and Sanat Kumar Mohapatra, our Director of Product Engineering got them deployed across the codebase. Not a wiki. Not a Confluence page. Markdown files at the root of the repo, next to the code. Each one covers who the customer is, what the product philosophy is, what we never compromise on, and what our AI architecture looks like. When an AI coding assistant opens a file in your repo, it has no context. It does not know who your customer is, what your moat is, or what you never compromise on. So it writes perfectly functional code that slowly drifts from your product philosophy in ways that are hard to see and harder to reverse. The product principles file at the root of the repo is the answer. One shared file covering who we serve, why we exist, and what good product looks like at Aurigo. Four product files covering each of our product lines. Every engineer reads the shared file when they join. Every AI coding assistant reads both before touching the code. Three things I learned writing them. Writing it down forces decisions you have been avoiding. You cannot write “our moat is twenty years of domain data and deep domain expertise encoded into our agents” without confronting whether your team actually builds that way every day. A principles file is not documentation. Documentation describes what the system does. A principles file describes why every decision was made and what you never compromise on. Those belong in different places. And the file from the founder lands differently than a wiki page from a product manager. Engineers know when the philosophy comes from the top. One gets read. The other gets ignored. If you are building an AI-native company and you have not written this file yet, do it today. Not for your engineers. For the AI coding assistants working alongside them who have no other way to understand what you are building and why. We build software that builds the world. Every line of code should know that.

83

5 Comments

"Every engineer is a Data/ML Engineer... if the Infra is right!" 🚀 Srikanth Sundarrajan's opening slide was a mic-drop moment that had every Platform and Infra engineer in the room (including me!) nodding in agreement. It perfectly captures our own (Anyscale) core mission: to build infrastructure so seamless it empowers everyone to innovate with data, AI & ML. InMobi's Encore'25 provided a fantastic look under the hood of their AI/ML infrastructure. For anyone obsessed with building and scaling machine learning systems, this was gold. Three things that stood out: Cost-Effective Model Training: Deep dive into their strategies for optimizing large-scale training jobs, intelligently leveraging spot instances and right-sizing resources without compromising on model development velocity. Real-Time Feature Stores: A key focus was their approach to building and maintaining a feature store that serves up-to-date data with single-digit millisecond latency, which is critical for their real-time bidding (RTB) models. The Rise of GenAI Infra: It was exciting to hear about their early infrastructure bets for supporting Generative AI and LLMs, particularly for dynamic creative optimization. This signals a major shift in the computational demands for the next generation of AdTech. Kudos to the InMobi engineering teams for their transparency and for building a truly impressive, high-performance ML platform. #AI #ML #Infrastructure #Encore25 #InMobi #MLOps #FeatureStore #GenAI #CloudComputing #AdTech

277

5 Comments

It has been an “eventful” week for many in production engineering. As I was reading the detailed post-incident report on the Amazon DynamoDB outage and related service disruptions (https://lnkd.in/gx9m779v) last night, I was reminded of the intricate dependencies that exist within and across modern systems — and how vast their blast radius can be. From Eight Sleep mattresses to Postman, Premier League games, and even Venmo — this truly was the shot heard around the world! It’s also a sobering reminder that production engineering is serious work. When incidents happen, the pressure to restore normalcy is immense. There’s no such thing as “vibe debugging” or “vibe loops” — let’s save that hype for hobby projects, not enterprise-grade systems in production. To all the #SRE, incident responders, and on-call engineers who worked tirelessly to get systems back online and communicate with customers with grace — thank you! #Reliability isn’t just about uptime — it’s about trust, teamwork, and the people who hold it all together. Here’s to those who keep the digital world running while most of us sleep 🙏

111

5 Comments

ACTORS + AI: Director/Actor/Producer/Writer AI Chef 🧑‍🍳 Hans Obma visits the Cozy AI Kitchen 🎂 https://lnkd.in/gjiAvEfP to share his insights on how AI has transformed his creative processes. You may have seen Chef Hans in such 📺 hits as Better Call Saul, Wandavision, or Narcos — and yes, he's probably the most handsome AI 🧑‍🍳 chef we've had in the kitchen to date :-). Hans needed to learn a new language ... how did he do it? He got multimodal AI models to help him speak Welsh. Hans needed a way to scale how he communicated with his stakeholders ... how did he do it? He created an authentic-to-him way of scaling by creating an instant entourage of supporting agents. HT 🧑‍🍳 Ross Heise 🧑‍🍳 Matt Scholz and the power of the Wisconsin network for bringing such a storied talent to the Cozy AI Kitchen! --- This 🍰 episode: https://lnkd.in/gjiAvEfP All 50 episodes: https://lnkd.in/g6upvbGX

89

11 Comments

The FAANG → startup gap is widening. When I was at Uber, we could still reliably hire great engineers out of FAANG. The velocity difference existed, but it was bridgeable. Now? For a seed or Series A team, the ramp is often too steep: ambiguity everywhere no guardrails shipping is the strategy “ownership” means end-to-end, not a slice Here’s what I think happens next as big tech layoffs continue: If you’re hiring: expect more noise. A larger pool of “available” talent won’t automatically translate to startup-ready talent. Your filter gets harder: slope > pedigree. If you’re a FAANG engineer: if you want risk, speed, and real ownership, jump earlier. After ~3+ years in a highly structured environment, the pivot gets meaningfully harder. Not because you’re not capable, but because the operating system is different. And I’m seeing a pattern: A lot of FAANG engineers want to move faster… But many early-stage startups aren’t prioritizing candidates who’ve spent 5+ years in big tech. Not a knock. Just a market reality. Curious if others are seeing the same thing: What’s your “threshold” where the FAANG → seed jump gets tough? 2 years? 4 years? More?

261

53 Comments