simdjzon-rpctravisstaloch/simdjzon-rpc
MIT
A fast json-rpc implementation in Zig based on simdjzon
7700
1
zig-package
build.zig.zonbuild.zig
View package on GitHubDependencies
Namesimdjzon
Commit
masterREADME
simdjzon-rpc
:warning: still in early development. :warning:
A fast json-rpc implementation in Zig based on simdjzon
Also includes a very competitive std-json-rpc implementation. See benchmarks below.
usage
$ zig fetch --save=simdjzon-rpc git+https://github.com/travisstaloch/simdjzon-rpc#main
// build.zig
// simdjzon-rpc
const simdjzon_rpc_dep = b.dependency(
"simdjzon-rpc",
.{ .target = target, .optimize = optimize },
);
my_exe.addModule("simdjzon-rpc", simdjzon_rpc_dep.module());
// std-json-rpc
const stdjson_rpc_dep = b.dependency(
"std-json-rpc",
.{ .target = target, .optimize = optimize },
);
my_exe.addModule("std-json-rpc", stdjson_rpc_dep.module());
// main.zig
const simjzon_rpc = @import("simdjzon-rpc");
const stdjson_rpc = @import("std-json-rpc");
examples
Echo Server
run and test http-echo-server
$ zig build http-echo-server
listening on http://127.0.0.1:4000
in another terminal
$ cd examples
$ python3 test.py
ok
Benchmarks
There are 2 benchmark scripts bench and bench-std-json. They are very similar. Their engines each have 4 rpc methods and randomly choose an input json from common 'test_cases_2'.
The following output can be reproduced by running bench.sh.
benchmark related build steps and options
$ zig build -h
#...
Steps:
#...
bench Run bench
bench-std-json Run bench-std-json
#...
Project-Specific Options:
#...
-Dbench-iterations=[int] number times for benchmark to loop. default 100.
-Dbench-iterations=[int] number times for benchmark to loop. default 100.
-Dbench-summary=[bool] whether or not to show bench timing and memory usage summary. default false.
-Dbench-use-gpa=[bool] whether or not to use zig's general purpose allocator. use std.heap.c_allocator when false. default false.
-Dbench-validate=[bool] whether to check that rpc output matches expected output. default false.
...
single benchmark summary
$ zig build bench -Dbench-iterations=30000 -Doptimize=ReleaseFast -Dbench-summary -Dbench-use-gpa=false
(simdjzon-rpc)
reqs=30000
time=16.066ms
req/s=1867.2K
-- CountingAllocator summary --
total_bytes_allocated 134.2MiB
bytes_in_use 17.8KiB
max_bytes_in_use 23.6KiB
allocs 19.06us/13 (avg 1.466us)
frees 430ns/7 (avg 61ns)
shrinks 622.401us/30000 (avg 20ns)
expands 2.355ms/59997 (avg 39ns)
total_time 2.997ms
benchmark results
Below simdjzon-rpc to std-json-rpc with these options
| build mode | allocator |
|---|---|
| ReleaseFast, ReleaseSafe | Gpa, C allocator |
These benchmarks happen at the end of scripts/bench.sh.
TLDR: simdjzon-rpc is between 6%-27% faster in 3/4 cases and 5.5% slower with the ReleaseSafe/Gpa combo. simdjzon-rpc runs 27% faster with a ReleaseFast, C Allocator combo.
ReleaseFast, C allocator
++ zig build -Doptimize=ReleaseFast -Dbench-use-gpa=false -Dbench-iterations=30000
++ ../poop/zig-out/bin/poop -d 2000 zig-out/bin/bench-std-json zig-out/bin/bench
Benchmark 1 (93 runs): zig-out/bin/bench-std-json
measurement mean ± σ min … max outliers delta
wall_time 21.2ms ± 293us 20.8ms … 22.4ms 3 ( 3%) 0%
peak_rss 1.53MB ± 61.6KB 1.44MB … 1.57MB 0 ( 0%) 0%
cpu_cycles 95.0M ± 906K 93.7M … 98.4M 3 ( 3%) 0%
instructions 274M ± 31.0K 274M … 274M 2 ( 2%) 0%
cache_references 240K ± 109K 185K … 747K 14 (15%) 0%
cache_misses 6.10K ± 436 5.66K … 9.10K 2 ( 2%) 0%
branch_misses 153K ± 2.25K 149K … 162K 2 ( 2%) 0%
Benchmark 2 (131 runs): zig-out/bin/bench
measurement mean ± σ min … max outliers delta
wall_time 15.3ms ± 348us 15.0ms … 16.8ms 9 ( 7%) ⚡- 27.8% ± 0.4%
peak_rss 1.44MB ± 25.2KB 1.31MB … 1.44MB 5 ( 4%) ⚡- 6.1% ± 0.8%
cpu_cycles 67.6M ± 594K 66.9M … 69.9M 10 ( 8%) ⚡- 28.9% ± 0.2%
instructions 180M ± 36.5K 180M … 180M 3 ( 2%) ⚡- 34.4% ± 0.0%
cache_references 38.9K ± 54.7K 21.8K … 406K 18 (14%) ⚡- 83.8% ± 9.1%
cache_misses 5.58K ± 191 5.23K … 6.35K 2 ( 2%) ⚡- 8.5% ± 1.4%
branch_misses 56.5K ± 526 55.5K … 59.3K 4 ( 3%) ⚡- 63.0% ± 0.3%
ReleaseFast, Gpa
++ zig build -Doptimize=ReleaseFast -Dbench-use-gpa -Dbench-iterations=30000
++ ../poop/zig-out/bin/poop -d 2000 zig-out/bin/bench-std-json zig-out/bin/bench
Benchmark 1 (70 runs): zig-out/bin/bench-std-json
measurement mean ± σ min … max outliers delta
wall_time 28.9ms ± 346us 28.4ms … 30.2ms 3 ( 4%) 0%
peak_rss 1.84MB ± 0 1.84MB … 1.84MB 0 ( 0%) 0%
cpu_cycles 129M ± 877K 127M … 131M 0 ( 0%) 0%
instructions 319M ± 74.8K 319M … 319M 3 ( 4%) 0%
cache_references 634K ± 110K 561K … 1.46M 4 ( 6%) 0%
cache_misses 6.90K ± 739 5.87K … 10.0K 4 ( 6%) 0%
branch_misses 159K ± 2.70K 155K … 164K 0 ( 0%) 0%
Benchmark 2 (75 runs): zig-out/bin/bench
measurement mean ± σ min … max outliers delta
wall_time 26.7ms ± 292us 26.2ms … 27.7ms 3 ( 4%) ⚡- 7.5% ± 0.4%
peak_rss 1.70MB ± 0 1.70MB … 1.70MB 0 ( 0%) ⚡- 7.1% ± 0.0%
cpu_cycles 119M ± 1.03M 118M … 122M 3 ( 4%) ⚡- 7.5% ± 0.2%
instructions 246M ± 94.8K 246M … 246M 4 ( 5%) ⚡- 22.9% ± 0.0%
cache_references 318K ± 157K 149K … 1.08M 2 ( 3%) ⚡- 49.8% ± 7.0%
cache_misses 5.89K ± 320 5.44K … 7.10K 5 ( 7%) ⚡- 14.6% ± 2.7%
branch_misses 58.6K ± 3.50K 57.0K … 87.7K 2 ( 3%) ⚡- 63.2% ± 0.6%
ReleaseSafe, C allocator
++ zig build -Doptimize=ReleaseSafe -Dbench-use-gpa=false -Dbench-iterations=30000
++ ../poop/zig-out/bin/poop -d 2000 zig-out/bin/bench-std-json zig-out/bin/bench
Benchmark 1 (73 runs): zig-out/bin/bench-std-json
measurement mean ± σ min … max outliers delta
wall_time 26.9ms ± 332us 26.3ms … 28.3ms 6 ( 8%) 0%
peak_rss 1.57MB ± 37.7KB 1.44MB … 1.70MB 6 ( 8%) 0%
cpu_cycles 121M ± 873K 119M … 124M 2 ( 3%) 0%
instructions 367M ± 29.2K 367M … 368M 1 ( 1%) 0%
cache_references 1.06M ± 66.0K 997K … 1.38M 5 ( 7%) 0%
cache_misses 6.33K ± 323 5.78K … 7.01K 0 ( 0%) 0%
branch_misses 141K ± 1.96K 137K … 147K 0 ( 0%) 0%
Benchmark 2 (80 runs): zig-out/bin/bench
measurement mean ± σ min … max outliers delta
wall_time 25.2ms ± 366us 24.8ms … 27.0ms 5 ( 6%) ⚡- 6.2% ± 0.4%
peak_rss 1.57MB ± 0 1.57MB … 1.57MB 0 ( 0%) + 0.2% ± 0.5%
cpu_cycles 113M ± 650K 113M … 116M 3 ( 4%) ⚡- 6.4% ± 0.2%
instructions 291M ± 27.8K 291M … 291M 2 ( 3%) ⚡- 20.9% ± 0.0%
cache_references 370K ± 98.0K 323K … 1.06M 9 (11%) ⚡- 65.2% ± 2.5%
cache_misses 6.00K ± 365 5.35K … 6.84K 0 ( 0%) ⚡- 5.2% ± 1.7%
branch_misses 61.5K ± 991 60.6K … 66.7K 5 ( 6%) ⚡- 56.6% ± 0.3%
ReleaseSafe, Gpa
++ zig build -Doptimize=ReleaseSafe -Dbench-use-gpa -Dbench-iterations=30000
++ ../poop/zig-out/bin/poop -d 2000 zig-out/bin/bench-std-json zig-out/bin/bench
Benchmark 1 (52 runs): zig-out/bin/bench-std-json
measurement mean ± σ min … max outliers delta
wall_time 38.6ms ± 580us 38.1ms … 41.1ms 8 (15%) 0%
peak_rss 1.93MB ± 57.3KB 1.84MB … 1.97MB 13 (25%) 0%
cpu_cycles 174M ± 655K 172M … 176M 3 ( 6%) 0%
instructions 448M ± 53.6K 448M … 448M 2 ( 4%) 0%
cache_references 1.92M ± 458K 1.58M … 4.00M 6 (12%) 0%
cache_misses 8.34K ± 890 6.89K … 10.9K 2 ( 4%) 0%
branch_misses 223K ± 4.00K 214K … 234K 0 ( 0%) 0%
Benchmark 2 (50 runs): zig-out/bin/bench
measurement mean ± σ min … max outliers delta
wall_time 40.7ms ± 407us 40.2ms … 42.0ms 3 ( 6%) 💩+ 5.4% ± 0.5%
peak_rss 1.97MB ± 0 1.97MB … 1.97MB 0 ( 0%) + 1.7% ± 0.8%
cpu_cycles 184M ± 1.52M 182M … 190M 3 ( 6%) 💩+ 5.9% ± 0.3%
instructions 404M ± 74.7K 404M … 404M 4 ( 8%) ⚡- 9.9% ± 0.0%
cache_references 978K ± 272K 721K … 2.19M 3 ( 6%) ⚡- 49.1% ± 7.7%
cache_misses 7.27K ± 592 6.42K … 9.18K 2 ( 4%) ⚡- 12.8% ± 3.6%
branch_misses 90.7K ± 4.30K 79.9K … 104K 6 (12%) ⚡- 59.3% ± 0.7%
notes
- adapted from ucall
- json-rpc spec
- related zig-json-rpc