#647 – October 12, 2025
When done right it cuts loading times
A complete guide to HTTP caching
44 minutes by Jono Alderson
Caching is the invisible backbone that makes websites fast and affordable to run. When done right, it cuts loading times and helps sites survive traffic spikes. When done wrong, it leaves sites slow and fragile. Many developers misunderstand basic caching rules. They confuse "no-cache" with "don't cache" or use "no-store" too often. These mistakes waste money and hurt performance.
P99 CONF: 60+ Tech Talks on Low Latency Engineering (Free + Virtual)
sponsored by ScyllaDB
P99 CONF is a free + virtual conference on low-latency engineering. AI/ML, Rust, Go, eBPF, distributed databases… it’s all on the agenda. Learn from experts at NVIDIA, Gemini, Meta, Disney, Uber, Pinterest, Rivian, Netflix, PayPal, & more at this highly technical (and highly interactive) conference. Discover optimization strategies and debate what’s next for performance. Get your free conference pass and 30-day O’Reilly book access too.
Beyond indexes: How open table formats optimize query performance
25 minutes by Jack Vanlightly
Jack explores why traditional secondary indexes from OLTP systems don’t translate well to modern open table formats like Apache Iceberg and Delta Lake. While relational databases rely on B-tree indexes for point lookups and selective queries, open table formats focus on reducing I/O through data layout strategies—partitioning, sorting, and metadata like column statistics and Bloom filters. Using Iceberg as a case study, Jack shows how query performance in analytical systems depends on data skipping, not row-level access.
Inside NVIDIA GPUs: Anatomy of high performance matmul kernels
about 1 hour by Aleksa Gordić
NVIDIA GPU matrix multiplication kernels use specialized hardware features to achieve peak performance. Modern transformers spend most compute time in matrix operations, making optimized GPU kernels crucial for efficiency. Aleksa covers GPU architecture fundamentals like memory hierarchy and compute units. He then builds up to state of the art techniques using tensor cores, asynchronous memory operations, and advanced scheduling patterns like Hilbert curves.
Tips for probabilistic software
6 minutes by Jason Liu
Jason offers practical tips for building probabilistic software, emphasizing experimentation, metrics, and decision-making under uncertainty. He encourages junior engineers to shift from deterministic thinking to outcome-driven iteration, treating even negative results as valuable insights.
The QNX operating system
39 minutes by Bradford Morgan White
Gordon Bell and Dan Dodge created QNX after working on a real-time operating system called Thoth at the University of Waterloo in 1979. They founded Quantum Software Systems in 1980 and developed QUNIX, later renamed QNX due to trademark issues with AT&T. QNX became a microkernel real-time operating system known for its reliability and network transparency. The system found success in automotive applications, embedded systems, and mission-critical environments, eventually being acquired by BlackBerry and continuing to power millions of vehicles today.
And the most popular article from the last issue was: