fork/rust_serialization_benchmark
git clone https://git.julia.blue/fork/rust_serialization_benchmark
c26f391
master
files
| .cargo/ | |
| src/ | |
| .gitignore | 47 B |
| Cargo.toml | 631 B |
| README.md | 1.9 KiB |
| Trunk.prod.toml | 224 B |
| build.rs | 1.5 KiB |
| index.html | 300 B |
Rust Serialization Benchmark Interactive Site
Inputs
- Bandwidth: in terabytes per month. 1 TB/Mo is 0.38 megabytes per second or 3.04 megabits per second
- CPU: fraction of CPU benchmarks were run on available for use (if > 1 assumes 0 overhead for parallelization)
- Dataset: (see ../README.md) changes messages/s to e.g. logs/s
- log: logs (benchmark size divided by 10000, equal to individual logs in benchmark)
- mesh: meshes (benchmark size)
- minecraft_savedata: saves (benchmark size divided by 500, equal to individual player saves in benchmark)
- mk48: updates (benchmark size divided by 1000, equal to individual updates in benchmark)
- Mode:
- serialize: Bandwidth usage is size of compressed data, CPU usage is serialization + compression
- deserialize: Bandwidth usage is size of compressed data, CPU usage is decompression + deserialization (crates without deserialize are excluded)
- round trip: Bandwidth/CPU usage is sum of Mode serialize and Mode deserialize (crates without deserialize are excluded)
- zlib: allow using zlib as Compression
- zstd: allow using zstd as Compression
Outputs
- Crate: which crate is being used for serialization/deserialization
- Compression: which compression algorithm is deemed the best (most messages/s) for that crate
- messages/s: how many messages could theoretically be sent per second based on available Bandwidth/CPU consumed by compressed data/serialization + compression
- Relative: normalized messages/s
- Bottleneck: whether Bandwidth or CPU runs out first (limiting messages/s)
Assumptions
- zlib/zstd have a constant speed irrelevant of Dataset (hopefully we can fix this)
- 1 message of size 1000 takes the same Bandwidth/CPU as 1000 messages of size 1
- The amount of messages that need to be sent per second is constant (if each day you had all of your messages in a 1-hour interval, your real CPU requirement would be 24x)