benchmarks's Introduction

Benchmarks

Kafka

Scenario

Message source: produced as fast as possible in parallel to the actual application.

Application: Read X msgs, create a protobuf struct and produce it to another topic. Commit automatically. Currently 1 thread for consumer, 1 for producer. Approximates at least one use case I was thinking about using Rust for.

.fresh.sh cleans and starts up a fresh kafka

measure-xx.sh starts up the message production and the application and prints time / mem taken at the end

Result

Conclusion

Rust takes 10% less CPU
Rust takes roughly the same time
Rust seems to take a little more memory, however, the memory is much stabler - e.g. Min (Go: 400mb, Rust: 560mb), Max (Go: 1260mb, Rust: 760mb)

Notes

Memory usage depends on how many messages need to be buffered - either currently consuming or producing.

Go is all over the place, 500-1200mb, GC?

Setting linger.ms to 0 for Rust increased the mem usage of Rust by like 200mb, network overhead?

Go CPU is around 88-100, Rust is around 80-85.

Metrics

Command	Metric	Min	Max	Mean	Median
go-single	userTime	191.44	213.77	196.73	193.16
rust-single	userTime	132.84	147.51	136.37	135.56
go-single	systemTime	74.77	85.10	77.42	76.10
rust-single	systemTime	86.20	97.16	88.83	88.16
go-single	percentCPU	99.00	105.00	103.92	105.00
rust-single	percentCPU	84.00	89.00	87.46	88.00
go-single	elapsedTime	254.87	298.98	262.52	256.70
rust-single	elapsedTime	249.74	290.76	255.76	252.77
go-single	maxResidentSetSize	402192.00	1260940.00	768365.85	725124.00
rust-single	maxResidentSetSize	569596.00	763076.00	720986.77	729944.00

Recommend Projects