Anirudh Rowjee @ rowjee.com

@AnirudhRowjee

LSM Trees @couchbase 🛠️ @bengalurusys @hsp_ecc • Databases, Dist Sys, Inference, Rust • runner, metalhead, poet⚡dare mighty things •
autodidact • views mine

Bengaluru rowjee.com Joined January 2021

3.3K Following 1.7K Followers 1.3K Posts

Latest posts 5

Pinned post

Anirudh Rowjee @ rowjee.com @AnirudhRowjee Aug 1, 2022

MapReduce is one of the foundational paradigms of Big Data processing, pioneered by Google!

In this blog post, I review the original paper, and add some diagrams to make it easy to understand!

rowjee.com/blog/papers/ma…

Anirudh Rowjee @ rowjee.com @AnirudhRowjee 2d

It’s honestly amazing that this is free - massive thanks to @1a1a11a and team!

Juncheng Yang Verified account @1a1a11a 2d

We just open-sourced PrefixSim — a fast simulator for KV-block prefix caches in LLM serving, along with two LLM serving datasets.

It lets you replay real production traces and compare 40+ eviction policies by the metric that actually matters: compute savings (prefill work avoided), not just hit ratio.

github.com/cacheMon/Prefi…

#LLM #PrefixCaching #Systems #OpenSource
View quoted post on X

Anirudh Rowjee @ rowjee.com @AnirudhRowjee 3d

Time to go all in on inference. What resources does everyone recommend?

Anirudh Rowjee @ rowjee.com @AnirudhRowjee Oct 26

Reaching out to anyone who's interested in/researching/building in LLM Inference - we'd love to hear you speak at our upcoming Special Edition on Inference at Bengaluru Systems on October 31st at @lossfunk

Bengaluru Systems (fka Bengaluru Systems Meetup) @BengaluruSys Oct 24

We’re back for our 23rd Bengaluru Systems meetup, this time on Inference with @lossfunk on October 31st!

Registration is now open; let’s talk systems!
hasgeek.com/bengalurusyste…
View quoted post on X

Anirudh Rowjee @ rowjee.com @AnirudhRowjee May 27

I'm going to be at @sigmod in Bengaluru next week. Looking forward to all the sessions on distributed storage, LSM Trees, storage engines, cloud native data systems - also lots of cool work around filters, consensus protocols, agentic workloads
happy to chat if you're there too!