On Thu, May 01, 2008 at 01:55:40AM -0700, Mika Borner wrote:
> Hi
>
> We are hosting a ruby on rails application for a customer. We first run the
> application on a T2000 and single thread performance sucked.
>
> Because of tight schedules we moved the app to an X2100, and performance
> seems ok so far.
>
> Anyway, I'm trying to find out where the bottleneck is, and run a dtrace to
> count the system calls for the mongrel process.
>
> It seems there are a lot of setcontext/nanosleep syscalls when issuing an
> http request. Can anybody explain me shortly, what these system calls do? I
> tried to google them up, but couldn't find an explanation that I'mable to
> understand :-)
>
> dtrace -n 'syscall:::entry/execname == "mongrel_rails"/ { @[probefunc] =
> count() }'
> dtrace: description 'syscall:::entry' matched 233 probes
> ^C
>
Change your program to something like:
syscall::setcontext:entry
/execname == "mongrel_rails"/
{
@[ustack()] = count();
}
and let's see what is causing that. The likely causes are either
that it's raining signals, or mongrel is co-routine crazy.
Either way setcontext() is quite an expensive thing to do
and this will likely lead to something useful to optimize.
-Mike
--
Mike Shapiro, Sun Microsystems Fishworks. blogs.sun.com/mws/
_______________________________________________
dtrace-discuss mailing list
[email protected]
