Perl's autoflush() and Gearman workers

[Brad Fitzpatrick]

Please!

On Tue, Mar 11, 2008 at 1:32 PM, Chris Goffinet <goffinet at yahoo-inc.com>
wrote:

> Brad,
> There was an issue I mentioned to hachi about the current implementation
> with Gearman (he managed to reproduce what I was seeing). It involves when
> you have a large number of workers (>100) connected to a gearmand server. It
> adds extra latency and grows as you add more generally. Its because the
> workers are constantly pounding the server asking for jobs when lots of
> workers are idle.
>
> I have the C version of gearmand that a person helped convert. I can post
> if you guys want it.
>
> On Mar 11, 2008, at 1:18 PM, Brad Fitzpatrick wrote:
>
> Not that I know of!  :-)
>
> I'm surprised to hear gearmand is anybody's bottleneck, though.  And if it
> is, you can just add more.  It scales out linearly.
>
> I assume everybody's on a modern Linux with epoll?
>
> On Tue, Mar 11, 2008 at 1:14 PM, Matt Sergeant <matt at sergeant.org> wrote:
>
> > On 9-Jan-08, at 3:52 AM, Chris Goffinet wrote:
> >
> > > :) Before I spend lots of time profiling, analyzing whats the
> > > bottleneck in gearmand (perl) I'd like to spend 5 minutes compiling
> > > gearmand c version to see if I notice anything different speed
> > > wise. Than at least I have something to compare profile wise to see
> > > where things are going slower.
> >
> > There's a C version of gearmand somewhere?
> > _______________________________________________
> > Gearman mailing list
> > Gearman at lists.danga.com
> > http://lists.danga.com/mailman/listinfo/gearman
> >
>
>
>
-------------- next part --------------
An HTML attachment was scrubbed...
URL: http://lists.danga.com/pipermail/gearman/attachments/20080311/63527f99/attachment.html