How OpenAI ingests petabytes of logs every day
In this blog we will see where OpenAI's logs go, and how one line of code saved its logging system. In June 2025, ClickHouse wrote about OpenAI's logs.
Topic
#performance
17 篇笔记
There's no reason for software to be slow anymore
📝 摘要 Dan Luu 论证 AI 编码智能体（coding agents）已将软件性能优化的成本降低了多个数量级——过去需要专业性能工程师数天/数周完成的工作，现在普通人输入几句话即可完成。文章通过多个案例展示这一趋势：FRE 正则引擎的 AOT 编译器切换为 ripgrep 带来 7% 提速；Azul 游戏 AI 借助 LLM 实现多线程和 10-20 项手工难以完成的优化，远超人工编写的…
Agentic Kernels in Production
TL;DR: We’ve built an agentic kernel development framework that identifies model-level optimization opportunities, generates improved kernel...
AI Agent 沙箱怎么选？我们把 runc、vArmor 和 Kata 跑了 30 轮
在 104 线程 Linux 物理机上实测 runc、runc + vArmor 与 Kata QEMU：Kata 用近一倍启动时间和数百 MiB 内存换独立内核，vArmor 则以接近 runc 的成本收紧行为边界。
io_uring：Linux 的高效异步 I/O 框架
iouring：Linux 的高效异步 I/O 框架 原文: Efficient IO with iouring — Jens Axboe, 2019 作者: Jens Axboe（Linux 块设备层与 I/O 子系统核心维护者） 说明: 本文翻译并梳理了 iouring 的设计思想、核心机制与使用方式。 📝 摘要 iouring 是 Jens Axboe 于 2019 年提出的 Linux…
Profiling LLM Inference with SGLang and Torch Profiler
This blog is my attempt to learn how to profile an llm served using sglang and its in built torch profiler integration that comes out of box...
Concurrency vs. Throughput: why more parallelism can make databases slower
Increasing throughput sometimes requires reducing parallelism. A look into why this is the case for Vitess + MySQL databases
What is WAL backpressure, and why does ClickHouse Managed Postgres need it?
How ClickHouse Managed Postgres uses WAL-aware backpressure to slow client writes, protect disk space, and let the archiver recover.
Why Kubernetes is the wrong tool for agentic workloads
At some point, everyone building ai infra wonders whether they should use Kubernetes. There’s no straight answer for it because it depends o...
Watching Go's new garbage collector move through the heap
Go 1.26 made Green Tea the default GC. We observe its cache-friendliness with perf, visualize the heap to see how Go allocates, and poke at a sparse-page problem a non-moving collector like Go's struggles with.
Highlights from Git 2.55
The open source Git project just released Git 2.55. Here is GitHub’s look at some of the most interesting features and changes introduced since last time.
Accelerating Gemini Nano models on Pixel with frozen Multi-Token Prediction
📝 摘要 Google Research 发布了一种将 Multi-Token Prediction (MTP) 增层改造（retrofit）到已冻结的 Gemini Nano v3 生产模型上的方法，在不修改基座模型权重的前提下，大幅加速 Pixel 设备上的端侧 LLM 推理。该方法已在 Pixel 9 和 10 系列实机部署，使 AI Notification Summaries、Proo…
Optimizing distributed AI inference: Advanced deployment patterns
Learn about the three optimization levers for distributed AI inference: prefill/decode disaggregation, KV cache strategy, and speculative decoding.
Migrating LiteLLM to Rust - Building the Fastest and Litest AI Gateway
LiteLLM is moving its AI gateway to Rust: 15x throughput, 11x less memory, and sub-1ms per-request overhead. No v2, no migration, your config stays the same.
Launching a million sandboxes per second in constant time
They say that in the future, everyone will have their own sandbox company for 15 minutes. Let's make that simpler, here's exactly how I buil...
Advanced Observability for Production Systems
From SLIs to LLM monitoring: 9 practices that separate mature teams from dashboard collectorsMost teams have dashboards. Few have observabil...
DuckDB Internals: Why is DuckDB Fast?
Walk through DuckDB's internals: how it skips serialization overhead, parses and optimizes SQL, and stores data in columnar row groups with zone maps.