On Wed, Jun 13, 2012 at 10:00 AM, Lauri Kasanen <[email protected]> wrote: > On Wed, 13 Jun 2012 09:55:45 -0600 > Eduardo Silva <[email protected]> wrote: > >> > I couldn't measure a difference due to the added epoll_del call: static >> > data both with and without gets the same 29k trans/sec. >> > >> > - Lauri >> >> i did some tests with valgrind/callgrind: >> >> the mk_sched_remove_client() perform the following: >> >> - without epoll delete = 0.38 >> - with epoll delete = 0.43 >> - conditional epoll delete OFF = 0.41 >> - conditional epoll delete ON = 0.46 >> >> comments ? > > The overhead of the conditional is too big, IMHO - it would be better to > always call. > > Some other random issues under high load may also have been due to this, even > without fork involved.
Yep, I think that the cheapest option is to always invoke epoll delete... (in a common scenario without fork there're not issues, the kernel take care of the epoll fd) regards, -- Eduardo Silva http://edsiper.linuxchile.cl http://www.monkey-project.com _______________________________________________ Monkey mailing list [email protected] http://lists.monkey-project.com/listinfo/monkey
