Bench signature verification

[metachris]

📝 Summary

Add a benchmark for signature verification

make bench outputs:

aws i3en.large - cpu: Intel(R) Xeon(R) Platinum 8175M CPU @ 2.50GHz

BenchmarkSignatureVerification-2   	     705	   1,735,334 ns/op

aws m1.metal

BenchmarkSignatureVerification-16    	     223	   5,377,079 ns/op

m1 max mbp - cpu: VirtualApple @ 2.50GHz

BenchmarkSignatureVerification-10            784           1,502,121 ns/op

Hetzner: cpu: AMD Ryzen 7 3700X 8-Core Processor

BenchmarkSignatureVerification-16    	     872	   1,313,259 ns/op

---

✅ I have run these commands

• make lint
• make test
• go mod tidy

[jtraglia]

LGTM. I'm not a fan of how go reports the results, but I don't think there's anything you can do about that. For example, I wish there were a table header. Without looking at the docs, I'm unsure what the first column is.

I also tested it on Linux/ARM64:

goos: linux
goarch: arm64
pkg: github.com/flashbots/go-boost-utils/types
BenchmarkSignatureVerification-8   	    1365	    872678 ns/op	    4929 B/op	      19 allocs/op
PASS
ok  	github.com/flashbots/go-boost-utils/types	1.290s

[metachris]

Thanks for the test! Yes, the output format is not great :(

The first column is the number of times the code has run. It seems on your ARM machine it was the fastest of all the test environments. What CPU is that?

[jtraglia]

Those results are from my Mac Studio (M1 Ultra) running a Linux VM with 8 CPU cores.

Here are some other results:

Mac Studio (M1 Ultra); macOS; go1.18.0:

BenchmarkSignatureVerification-20    	    1491	    804403 ns/op	    4933 B/op	      19 allocs/op

MacBook Air (M1); macOS; go1.18.1:

BenchmarkSignatureVerification-8   	    1471	    815220 ns/op	    4931 B/op	      19 allocs/op

Which makes sense. The M1 Ultra is essentially a scaled up base M1 processor. They use the same cores, but on M1 Pro/Max/Ultra performance cores are at a slightly higher clock (3.23 GHz vs 3.20 GHz) than the base M1 performance core. The results for MacBook Pro with M1 Max in your comment seem wrong. Which version of golang is being used there?

Also, it seems that this benchmark is only being run on a single core. Is that intended?