Re: [Libevent-users] HTTP client-side functions performance issues
On Wed, Jul 8, 2009 at 4:30 PM, Springande Ulv wrote: > Without having looked at your code I would guess the following. By > using the http part of libevent you do a lot of allocation and data > copying into buffers as well as function callbacks. While ab does no > such thing and write and read directly on the sockets. Also note that > there are minimal differences from select and other event mechanisms > for the small socket sets so there is not much to gain from using > epoll or kqueue in this case. You might want to try the benchmark tool with libevent-2.0.1-alpha or directly from trunk. There has been a lot of work to reduce data copies. Niels. ___ Libevent-users mailing list Libevent-users@monkey.org http://monkeymail.org/mailman/listinfo/libevent-users
Re: [Libevent-users] HTTP client-side functions performance issues
On 8. juli. 2009, at 20.17, Rocco Carbone wrote: > it the HTTP client-side part of the libevent so much performant > as the server-side? are the HTTP-client functions usable > for applications with strictly performances requirements? Without having looked at your code I would guess the following. By using the http part of libevent you do a lot of allocation and data copying into buffers as well as function callbacks. While ab does no such thing and write and read directly on the sockets. Also note that there are minimal differences from select and other event mechanisms for the small socket sets so there is not much to gain from using epoll or kqueue in this case. ___ Libevent-users mailing list Libevent-users@monkey.org http://monkeymail.org/mailman/listinfo/libevent-users
Re: [Libevent-users] HTTP client-side functions performance issues
On Wed, Jul 08, 2009, Rocco Carbone wrote: > You can see that 'cb' is as fast as 'ab', while 'lb' > is slower than both of them, contrary to all my expectations. "cb" will be hilariously faster when you: * add extra latency between client and server; and/or * bump the concurrency rate up to say 2000 As for performance - install and get familiar with oprofile. :) adrian ___ Libevent-users mailing list Libevent-users@monkey.org http://monkeymail.org/mailman/listinfo/libevent-users
Re: [Libevent-users] HTTP client-side functions performance issues
Rocco Carbone wrote: > > So my big question for you and the list is: > > it the HTTP client-side part of the libevent so much performant > as the server-side? are the HTTP-client functions usable > for applications with strictly performances requirements? > > > Is the CPU on the test driver system max'd out? Have you tried to run under either a system level profiler or the profiling mode of the valgrind package? ___ Libevent-users mailing list Libevent-users@monkey.org http://monkeymail.org/mailman/listinfo/libevent-users