This was a fun one! For @polarsignals.com's latest hackathon, I wrote an FSST decompression CUDA kernel and then optimized it by profiling with our new GPU PC sampling feature. www.polarsignals.com/blog/posts/2...

My Journey Optimizing a CUDA Kernel with Polar Signals | Polar Signals

A storage engineer's descent into warp stalls, memory throttles, and aligned stores

polarsignals.com

If you're doing any performance work on GPUs, this is huge. PC sampling is a powerful tool for GPU optimization, but causes serial kernel executions. We've released production-ready continuous PC sampling so you can get GPU performance insights on your production workloads.

Polar Signals@polarsignals.com · 2mo ago

🚀 Launching today: GPU PC sampling in production. Instruction-level profiling telling you why and where warps stall. Whether it is memory latency, scoreboard waits, or barriers. 🧊 Previously NSight-only territory. Now running in prod with low overhead. www.polarsignals.com/blog/posts/2...

New Blog Alert! Learn how we overcame eBPF limits to unwind deeply nested PyTorch training stacks and learned some cools things along the way: www.polarsignals.com/blog/posts/2...

Python Zebra Stacks | Polar Signals

Unwinding Deep PyTorch Training Stacks with eBPF

polarsignals.com

🤖Using Polar Signals Cloud and Claude Code to autonomously optimize your code every day! 📈 Also: Check out our new website! What do you think? www.polarsignals.com/blog/posts/2...

We've been working on a major schema migration at @polarsignals.com that has unlocked some pretty wild improvements. Apart from a 50% (!!) reduction in storage size thanks to logical compression in vortex.dev, there has been a dramatic improvement in query performance.

Tired of Node.js flamecharts full of `a`, `m`, and `_0x3f2`? We just shipped source maps support for Node.js profiles in Polar Signals Cloud. Just add a plugin to your build pipeline, and your profiles show the real code. Know more: www.polarsignals.com/blog/posts/2...

Do more with less. | Polar Signals

Polar Signals Cloud is an always-on, zero-instrumentation continuous profiling for CPU, GPU, and Memory that helps improve performance, understand incidents, and lower infrastructure costs.

polarsignals.com

I'm happy to share that I will be speaking at #QConLondon on March 16 on deterministic simulation testing at Polar Signals. For anyone interested in attending: SPEAKER60QUK26 for £60 off

Claude is down. I guess I have to learn how to code again.

📢 It’s a new year so we’re announcing something new which is the Sandwich view: A way to visualize a function’s callers and callees side by side for better debugging and performance work. Read more about it here: www.polarsignals.com/blog/posts/2...

Introducing the Sandwich View

A new way to visualize a function’s callers and callees side by side.

polarsignals.com

My EuroRust talk has been uploaded to YT! I cover how we architected our Rust database at @polarsignals.com as a set of state machines to enable deterministic simulation testing and catch hard-to-reproduce bugs before they hit prod.

EuroRust@eurorust.eu · 7mo ago

A new talk is out! 🙌 @asubiotto.com shows how deterministic simulation testing (DST) uses random seeds and state machines to explore countless execution paths—and reliably replay failures when things go wrong 🦀🎲 👉 youtu.be/V12hHoagv-E

A simple 50-line code change saves us 30% of CPU across our whole infrastructure. This is the magic of system-wide profiling with @polarsignals.com. Easy to spot, easy to fix.

performance[vortex-array]: don't call is_valid to count bytes in varbinview by asubiotto · Pull Request #5814 · vortex-data/vortex

I noticed that the better blocks compressor uses count_referenced_bytes which calls is_valid on each view and results in an expensive scalar_at call. This was 30% of our system-wide CPU usage over ...

github.com

🚀 We recently announced our NVIDIA CUDA always-on profiler. Check out this blog on how to try it out super quickly with your CUDA workload today!🧊 www.polarsignals.com/blog/posts/2...

Profiling NVIDIA CUDA in Kubernetes

The easiest way to get started Profiling CUDA in Kubernetes

polarsignals.com

Happy to share that I've been nominated to the Vortex Technical Steering Committee! It's been fun and productive switching to Vortex from Parquet as our storage format @polarsignals.com and I'm excited to continue contributing to the Vortex project. vortex.dev/blog/novembe...

🧊 New on the Polar Signals Blog — Our Delta Lake Fork Purpose-built for our continuous profiling product. In our latest post, we walk through how Delta Lake works, and the changes we've made to improve performance for our product. 👉 Read the full post: buff.ly/KwHINtO

Great Lakes: A Delta Lake Fork

Our updates to Delta Lake Table Format

polarsignals.com

Sometimes an interface is so ubiquitous that performance limitations inherent to its design are just accepted. This has been our experience with Parquet. We switched to Vortex which aligned much better with our needs and saw a 70% performance improvement across the board. Read more below 👇

Do more with less. | Polar Signals

Polar Signals Cloud is an always-on, zero-instrumentation continuous profiling for CPU, GPU, and Memory that helps improve performance, understand incidents, and lower infrastructure costs.

polarsignals.com

Ever wondered how your browser devtools jumps from bundle.min.js:1:27698 → src/index.ts:73:16? 🧊 We unpack the internals of JS Source Maps: Base64 VLQ encoding, mapping deltas, and how your original TS code gets reconstructed from minified bundles. 🔗 www.polarsignals.com/blog/posts/2...

The Inner Workings of JavaScript Source Maps

A deep dive into how JavaScript source maps work under the hood, with examples showing how all the pieces fit together.

polarsignals.com

Introducing the world's first open source continuous NVIDIA CUDA profiler. Zero code changes required, low overhead, and always on.

Polar Signals@polarsignals.com · 10mo ago

🚀 We're super excited to launch NVIDIA CUDA Profiling today! Check out the blog on how we made it happen! 🧊 www.polarsignals.com/blog/posts/2...

Took advantage of the lunch break to replay this great talk by Alfonso Subiotto (@asubiotto.com) making the case for a State Machines based approach to Deterministic Simulation Testing, being used at @polarsignals.com, very interesting! Their inspiration for this: sled.rs/simulation

We're excited to be at @rustconf.com 2025! Let's talk about profiling Rust with zero compromise! #rustconf25

We recently made some performance improvements to our Flame Graph by implementing backend pre-computation and viewport culling. Read more in our latest blog post: www.polarsignals.com/blog/posts/2...

Parallel Flame Graph Rendering

How we improved our flame graph rendering by moving computation to the backend.

polarsignals.com

I wrote about why we should treat German string encoding/StringViews as Just Another Encoding™

Polar Signals@polarsignals.com · 11mo ago

German strings 🇩🇪 have become very popular in the database world. However, they're not a silver bullet. Read more in our latest blog post:

Do any database systems "columnarize" the German string encoding so rather than storing a 16-byte view for each element which includes length+prefix+pointer you'd store all lengths together, prefixes together etc...? Seems like prefix operations would be more SIMD friendly.

The hardest part about writing a blog post is the introduction. Here I am, 4 hours later.

Tame the OOM killer with OOMProf, the new eBPF memory profiler for Go programs being taken out by the kernels OOM killer! www.polarsignals.com/blog/posts/2...

OOMProf - Profiling on the Brink

Using eBPF to record your programs dying breathes

polarsignals.com

We're excited to share that the Parca Agent now works offline! Store CPU profiles locally when disconnected, upload when ready. Because not all computing happens in data centers. Read how we built it: www.polarsignals.com/blog/posts/2...

Offline Mode for the Parca Agent

Collect and upload profiles -- even without a network connection

polarsignals.com

New storage format RFC in the works 👀 Early preview:

🚀 Today, we're excited to launch Off-CPU profiling! Check out our blog on what we needed to do to make it most useful and how you can get started with it! 🧊 www.polarsignals.com/blog/posts/2...

Introducing Off-CPU Profiling

How Off-CPU profiling works and how to get the most out of it

polarsignals.com

@metalmatze.de built an MCP server so you can hook up your agent to your profiling data. Check it out!

Matthias Loibl@metalmatze.de · last yr.

🔥 Have you ever stared at a flame graph wondering why your app is slow? Using the new @PolarSignalsIO MCP, you can now ask "Why is my service slow?" and e.g. @anthropic.com's Claude Code or @cursor.com.web.brid.gy will analyze your production profiles and suggest code fixes.

📔Writing our Rust database from scratch gave us a golden opportunity to implement deterministic simulation testing as a core feature. Learn about the architecture that gives us full control over DST ingredients. 🧊 www.polarsignals.com/blog/posts/2...

Deterministic Simulation Testing in Rust: A Theater Of State Machines

How we designed our database for complete control over concurrency, time, randomness, and failure injection.

polarsignals.com

Writing our Rust database from scratch gave us a golden opportunity to implement deterministic simulation testing as a core feature. I wrote a blog post about the architecture that gives us full control over concurrency, time, randomness, and failure injection. (link in comment)