Benchmark comparing read and write throughput

[markmandel]

What type of PR is this?

Uncomment only one /kind <> line, press enter to put that in a new line, and remove leading whitespace from that line:

/kind breaking
/kind bug
/kind cleanup
/kind documentation

/kind feature

/kind hotfix

What this PR does / Why we need it:

Wanted to be able to highlight if we had bottlenecks in performance on read vs write operations on the proxy.

This adds an extra benchmark to throughput.rs called "readwrite" and follows a similar pattern as the overall throughput benchmark, with both direct and proxies traffic utilised as extra comparison values.

Which issue(s) this PR fixes:

Work on #410

Special notes for your reviewer:

This is my first benchmark in Rust - feedback appreciated.

Would be interested if we think the new benchmark should be in a readwrite.rs ? Since it reused some machinery and was still essentially about throughput I figured it would make sense to stay in the same file, but wasn't 100% on it.

[markmandel]

Noting some interesting results from the benchmark. I've attached one of the readwrite runs as a PDF.
readwrite Summary - Criterion.rs.pdf

On average, throughput looks about the same:

Were I see some differences (and it's not 100% consistent, so we should triple check the results, and I should run it without other processes on the machine, for sure - but it does align with what I've seen previously on Xonotic+grafana graphs), is that the outliers on read tend to be much larger than that of write.

I'm cherry picking here, but to illustrate the general trend I'm seeing here is read/508 bytes:

And the corresponding write/508 bytes:

This can fluctuate between runs though, so I should run this in a cleaner environment and do multiple runs before definitely drawing conclusions.

The other thought I had too was it might be worth doing benchmarks with multiple udp clients (maybe more than current cpu on the machine?) and multiple servers, and look at the same things.

Thoughts?

[quilkin-bot]

Build Succeeded 🥳

Build Id: 9fe53ea8-bc27-47ab-bc16-2b98c6a9ea5a

To build this version:

git fetch [email protected]:googleforgames/quilkin.git pull/479/head:pr_479 && git checkout pr_479
cargo build

[XAMPPRocky]

Would be interested if we think the new benchmark should be in a readwrite.rs ?

I think it's fine for now, and maybe split it up once we have even more benchmarks.

The other thought I had too was it might be worth doing benchmarks with multiple udp clients (maybe more than current cpu on the machine?) and multiple servers, and look at the same things.

I think this would be a good variable to benchmark, another variable worth benchmarking in my opinion is the amount of filters in a given configuration. For example, how does it look with 5 filters vs 10? We'd probably just want to use the Pass filters for that because we really just want to benchmark the internal machinary rather the performance of an individual filter (though I could also adding benchmarks for specific filters as well [routing, compression, etc]).