Games
Should a game ship its audio as OGG, MP3 or AAC?
We encoded one exact-length music loop and ten sound effects every common way. Size barely separated the formats. Loop length did, and it depended on the decoder as much as the codec.
Blog / Topic
Performance
Benchmarks with the method shown — what is actually fast, what only looks fast, and by how much.
AI & Data
How accurate is HyperLogLog for counting unique visitors?
HyperLogLog met its quoted error at a million visitors, but drifted where it switches estimators and collapsed when a weak hash met sequential user IDs.
Games
What is the fastest way to compute field of view on a grid?
Recursive shadowcasting was the fastest field of view method that missed nothing. The version most people copy leaks through walls, and one line fixes it.
Games
Which procedural dungeon generation algorithm should I use?
Cellular automata caves were almost never fully connected. Repair them with a flood-fill tunnel, because discarding and retrying often failed.
Integration
Does HTTP keep-alive make API calls faster?
Reusing an HTTPS connection made each API call about 3x faster in our tests. It also opens a stale-socket race, and one idle-timeout setting closes it.
AI & Data
Which fuzzy matching algorithm finds duplicate customer names?
Trigram Jaccard on normalised names scored best of five algorithms. But no string score can tell apart two Vietnamese people who share a family and middle name.
Games
How long does it take to generate a terrain chunk with noise?
One noise chunk fits easily in a frame; a sprint across a chunk border can blow it. Simplex never beat branch-free Perlin, and a switch statement hid a trap.
Games
Is Math.random random enough for loot drops?
Math.random was as fair as crypto over 10 million loot rolls. The streaks players blame on the RNG are normal maths, and only a pity timer changes them.
Games
How much of an animation can you throw away?
About 97% of it. Constant-channel removal, keyframe reduction and quantisation, measured on a synthetic humanoid rig against joint error.
Infrastructure
Why did the first request take eight seconds?
A new subdomain took 8.1 seconds to first byte, then was fast. Inside the container, startup took under half a second. The rest was outside it.
Games
How much does sorting sprites by depth cost per frame?
Rebuilding the sprite list every frame made depth sorting up to 8x slower than re-sorting the list we kept, enough to blow a whole frame budget.
Games
Does object pooling actually make anything faster?
In our Node game-loop tests, the usual free-list pool slowed the median frame. Only a typed-array pool helped, and only frame-time spikes at high spawn rates.