Why could autocannon produce more requests than wrk?

[diorahman]

Thanks for providing such a great tool for benchmarking.

What is the underlying explanation on the capability of autocannon to produce so many requests? Is that because of its internal httpClient? I could find that autocannon does not use the standard http.request() to initiate the request, why is that?

Thanks!

[mcollina]

Because it uses a different concurrency model, and it supports HTTP pipelining. We use a custom HTTP client that allows us to avoid any data allocation to send a new request.

[diorahman]

Thanks, @mcollina!

[subomi]

@mcollina this is about event loops against threads?

[mcollina]

@Subomi yes, and autocannon supports HTTP/1.1 pipelining.

[subomi]

Thanks ..