`autocannon` is too slow to benchmark Deno or Bun correctly
Jarred-Sumner opened this issue · 2 comments
For many of the results on the homepage, you are benchmarking autocannon. When the HTTP client is slower than the HTTP server, you are benchmarking the HTTP client. That's why the numbers get so close towards the top -- they're within the margin of error for autocannon.
Deno and Bun's numbers will be more accurate (higher) using a tool that is better at keeping up. wrk is also not quite fast enough.
oha would be a more accurate choice, though http_load_test would be best.
Interesting! I tried to use a package called Bombardier but reverted back to autocannon because it provided better json output. I initially wanted to use Oha but that does not to my knowledge output json or any other machine-readable format?
Since the PR was reverted, could we reopen this issue?