extremely slow, slower than gnet by 40% benchmark because of syscall6

[hiqsociety]

possible to fix this?

"prefork mode" of hertz for "Hello World!" for techempower benchmark
https://github.com/cloudxaas/gosprapp/blob/main/web/app/app.go

possible to provide guidance on how to speed it up?
gnet ~ 95k req/s
hertz ~ 55k req/s

that's almost 40% slower than gnet

[li-jin-gou]

Can you provide detailed code and pressure test environment? I only see hertz's code.

[hiqsociety]

@li-jin-gou
the code is compared with
https://github.com/TechEmpower/FrameworkBenchmarks/blob/master/frameworks/Go/gnet/src/main.go

wrk -t 1 -d 10 http://.... :8080 / 8081

ubuntu 22.04, i5, 4 threads cpu core

pls try it out.
io_uring is the next step. just saying that's all. thx

[hiqsociety]

@li-jin-gou trying to use this framework with "prefork mode" to compete in techempower benchmark

do guide on how to make it faster. thx

[li-jin-gou]

This is pr in TechEmpower/FrameworkBenchmarks#8483 @hiqsociety We found that the code for the benchmark was not the same, which could lead to inaccurate results.
I don't think it makes much sense to compare gnet and hertz, maybe look at gnet and netpoll.

• netpoll-benchmark: https://github.com/cloudwego/netpoll-benchmark

[li-jin-gou]

I think hertz support for prefork could be a higher priority, as it seems like his performance gains are huge. @hiqsociety

[hiqsociety]

@li-jin-gou i dunno what this is about but hertz should be compared with gnet / fiber and it's still very slow compared to either.

whatever mode you prefer, just make it faster than what it is currently.

it's really too slow, defeats the purpose of using hertz.

This is pr in TechEmpower/FrameworkBenchmarks#8483 @hiqsociety We found that the code for the benchmark was not the same, which could lead to inaccurate results. I don't think it makes much sense to compare gnet and hertz, maybe look at gnet and netpoll.

• netpoll-benchmark: https://github.com/cloudwego/netpoll-benchmark

[hiqsociety]

@li-jin-gou it's fixed? possible to open this issue as a continual speed improvements update?
i just submitted a pull request to techempower benchmark on the prefork mode.

is there any guidance on built in prefork mode for hertz or can just use mine?

[li-jin-gou]

@li-jin-gou it's fixed? possible to open this issue as a continual speed improvements update? i just submitted a pull request to techempower benchmark on the prefork mode.

is there any guidance on built in prefork mode for hertz or can just use mine?

OK, #8483 already merged and @Skyenought will explore ways to optimize performance include Prefork etc in https://github.com/hertz-contrib/performance.

[hiqsociety]

this non-preforked results make hertz look so bad.
https://www.techempower.com/benchmarks/#hw=ph&test=plaintext&section=data-r22

my prefork version is accepted into the latest techempower benchmark repo. hope to be included in hertz as default prefork.

[li-jin-gou]

this non-preforked results make hertz look so bad. https://www.techempower.com/benchmarks/#hw=ph&test=plaintext&section=data-r22

my prefork version is accepted into the latest techempower benchmark repo. hope to be included in hertz as default prefork.

Don't really want to maintain prefork as there is no real user demand for it. If a user wants to use prefork, I think it would be a good idea to refer them to your code. TechEmpower/FrameworkBenchmarks#8528