3 ms·
And wrk2 (https://github.com/giltene/wrk2 https://github.com/giltene/wrk2) is the way to measure it.
by matt4077 10y ago
And wrk2 (https://github.com/giltene/wrk2 https://github.com/giltene/wrk2) is the way to measure it.
- Groxx 10y agoSo most benchmarking tools don't do this? Any idea if `ab`[1] does? I've been assuming that they all fired off requests at constant rates, 'cuz yeah, otherwise it's pretty inaccurate compared to real-world load. [1]: https://httpd.apache.org/docs/2.2/programs/ab.html https://httpd.apache.org/docs/2.2/programs/ab.html
- zzzcpan 10y ago`ab` doesn't. But then again, benchmarking http services for latency doesn't really work. It's too hard to reproduce very complex real-world environments to get any useful latency data out of it. Just record response times where possible instead.
- matt4077 10y agoI just checked wrt and as it turns out, they did indeed add it in the meantime. Wrt2 was about the only one when I need it a year or two ago. Most people here seem to be working with so much traffic that they primarily care about throughput, but I live of about 300 pageviews/day. With the way that people and google react to latency, it's a good day when I can shave 10ms of every request. (Can't wait for ruby->elixir transition to be complete. So far it seems to safe about 100ms).
- eggnet 10y agowrk has included corrections for coordinated omission. It is a different implementation, not sure what all of the differences are.