Topic
Everything filed under Performance, newest first.
RSS · JSON · All topics
Redis vs Valkey vs Dragonfly vs KiviDB
warefeats benchmarks Redis, Valkey, Dragonfly, and KiviDB 1.2.0 on a 64-core Graviton3, comparing throughput and RSS memory per key after a KiviDB storage rewrite.
9 min · 2,141 words
Tuist Runners: Going down to the bare metal on macOS and Linux
Managed macOS and Linux CI runners with colocated caching and insights down to the metal. Invite-only while we grow capacity.
10 min · 2,231 words
From Static to Polymorphic Inference: A new era for production open-source inference systems
Models, traffic and chips are constantly changing. Polymorphic Inference adapts to each, and Together Compiler uses AI agents to keep production inference tuned.
7 min · 1,647 words
benchfmt.rs: Bringing Go's Benchmarking and benchstat Workflow to Rust
Bartek Płotka introduces benchfmt.rs, a zero-dependency Rust benchmark runner and Criterion converter for Go's benchstat, and walks through why Go's benchmarking workflow works well, plus key=value projections, per-op heap allocation tracking and pprof profiling for Rust benchmarks.
21 min · 4,799 words
How we halved DNS build delay and cut memory use by 85%
Cloudflare's Alex Fattouche describes how the Zone Builder in its DNS propagation pipeline, now handling ~700 record changes per second, was reworked to do targeted zone builds instead of expensive full rebuilds, halving build delay and cutting memory use by 85%.
17 min · 3,893 words
How hard is it to check for whitespace?
Sergey Tselovalnikov follows Guava's surprisingly complex isWhitespace matcher down a rabbit hole into Unicode's many whitespace characters, how different languages define whitespace, and the perfect-hashing trick that makes the check fast.
6 min · 1,375 words
The AI cost of JavaScript frameworks
Nolan Lawson has coding agents build the same app five times in each of a dozen JavaScript frameworks, from Next.js and Angular to SvelteKit, Solid, htmx and vanilla Vite, and compares development time, tokens spent, lines of code and Lighthouse scores to argue for evidence-based framework choices.
16 min · 3,730 words
Making SIMD Code Scale to Any Vector Width
Stream HPC's Nemi Verploegh shows how to restructure vec3/vec4-style SIMD code, using a frustum culler as the example, so it scales from SSE to AVX and wider: turn vectors into lanes so eight dot products take three instructions, convert early-outs into masks, precompute splatted values the compiler cannot keep in registers, and write results a block at a time, for roughly 20-25x overall speedup.
19 min · 4,353 words
It's the memory latency, stupid
The Pivot blog explains what a database CPU at 100% is really doing, walking through memory stalls, CPU caches and perf counters, and shows how radix partitioning makes GROUP BY aggregations faster.
9 min · 2,159 words
Do We Still Have to Denormalize? Have Your Schema and Search It Too
Stu Hood explains how ParadeDB handles relevance-sorted search and facet aggregations across joined Postgres tables, so teams no longer need to denormalize their schema into a separate search index, with benchmarks against plain Postgres.
14 min · 3,118 words
Can GROUP BY still be optimized after all these years?
Pivot walks through how analytical databases execute GROUP BY and shares a novel optimization for GROUP BY queries with a Top-K operator above them, pruning partitions during execution for a 1.6x speedup on a ClickBench-style query.
6 min · 1,372 words
11 nanoseconds per word: Ridiculously Over-Optimizing Anagrams
David G. Andersen builds a solver for an anagram-enabled word ladder game and then over-optimizes its anagram indexing in Rust, using collision-free prime-product (Gödel numbering) identifiers, packed inputs and branchless loops to reach about 11 ns per word.
9 min · 2,171 words
Billion-query-per-second lookup tables in C++
Daniel Lemire shows how to replace std::unordered_map for small, compile-time-known string sets (HTTP methods, headers, keywords) with a constexpr perfect hash, cutting each lookup to a single comparison and about five cycles, and shares a single-header C++ library and benchmarks built with Francisco Geiman Thiesen of Microsoft.
5 min · 1,221 words
SIMD-accelerated computer vision on a $2 microcontroller
shranav describes writing a FAST corner feature detector for the ESP32-S3 using its little-known 128-bit SIMD instructions, working around misaligned reads and signed-only saturating arithmetic, and roughly doubling throughput to about 11.2 megapixels per second, around 6ms per QVGA frame.
4 min · 994 words
We should be more ambitious with the software we build
Dan Luu argues that AI coding tools let skilled engineers attempt far more ambitious software, using Michael Malis's pgrust (a Postgres rewrite with JIT-style optimizations whose testing has surfaced real Postgres bugs) as the running example, and explains why programming expertise still matters for judging model output.
27 min · 6,109 words
Memory is slow, Disk is fast - Part 3
Jared Hulbert benchmarks an on-disk sorted-set structure (DiskArray on 8 NVMe drives, 264 MiB RAM) against an in-RAM AVL tree at 2 billion keys and finds the disk-based design faster at search, rank, insert and delete, arguing Big O misleads and that memory must be treated like a parallel, pipelined block device.
19 min · 4,345 words
Go 1.28 HTTP/2 Parking: Measuring Memory Savings
Mohammad Abdolirad benchmarks the HTTP/2 connection parking change in Go 1.28 with 5,000 idle connections, finding the same server binary used 85 to 93 MiB less RSS with parking enabled, with fewer goroutines and less stack memory, and discusses why HeapAlloc told a less simple story and the limits of the test.
5 min · 1,204 words
Java's new AOT cache feature is in a rough state
Christopher Schnick of XPipe details the pitfalls of shipping the JDK's new ahead-of-time (AOT) cache in a real desktop application distributed to users: how training runs and cache creation work, conditions under which the cache is silently ignored, JVM crashes caused by CPU feature mismatches between build and user machines, and the build scripts XPipe now uses to roll it out safely.
22 min · 5,011 words
Xiangpeng Hao introduces bbtree, a zero-dependency Rust B-tree that he reports is 5x faster than std's BTreeMap on one thread and 12x faster than crossbeam-skiplist at 96 threads, explains how it cuts cache misses and coherence traffic, and describes how the LLM-written code is kept correct with human review, formal verification, deterministic simulation and fuzzing.
19 min · 4,484 words
OrgWebAlchemy: Benchmarking, Profiling, and the Long Road to Real Org Support
Josep Bigorra details how the Guile OrgWebAlchemy Org-mode parser gained Markdown output, figures, named elements and table rows, and got up to 6.4x faster after profiling uncovered a quadratic collapse pass and a per-character scanner.
26 min · 6,059 words