I guess the question to ask is: Where is the bottleneck? It's worth profiling a request to see where that 50.3ms is spent.
--Dan On Fri, Aug 17, 2012 at 5:38 PM, Ylan Segal <[email protected]> wrote: > Hi Guys, > > Is anyone using anything other than thin on heroku cedar? > > I was trying to see if any of the above web servers has a significantly > different performance, but I am finding it doesn't really seem to matter. > > I tried one of my real production apps (for a benchmark that is meaningful > to me) on a staging server and tried to measure performance. > > I tried: > > - Thin (as suggested by heroku). > - Unicorn (configuring for 3 workers) > - Puma (as configured by default). > > After deploying with each server I tried using httperf like so: > > httperf --hog --server $URL --num-conn 1000 --ra 20 --timeout 5 > > The weird part, is that it seems that the performance is the same for all > web servers: > > Thin: Request rate: 19.8 req/s (50.5 ms/req) > Unicorn: Request rate: 19.9 req/s (50.3 ms/req) > Puma:Request rate: 19.9 req/s (50.3 ms/req) > > Any ideas why I am not seeing a difference in the tests? I have several > theories, but would like to hear what the group has to say. > > -- > Ylan Segal > [email protected] > > -- > SD Ruby mailing list > [email protected] > http://groups.google.com/group/sdruby > -- SD Ruby mailing list [email protected] http://groups.google.com/group/sdruby
