httpd-dev mailing list archives

Site index · List index
Message view « Date » · « Thread »
Top « Date » · « Thread »
From Dean Gaudet <>
Subject Re: roy's l_c perf patch and spareservers
Date Sat, 15 Feb 1997 03:51:54 GMT
Oh yeah, once I have the patch from SGI for FIN_WAIT_2 timeouts I can test
the performance of Roy's patch... but right now I can't really afford
death by FIN_WAIT_2.


On Fri, 14 Feb 1997, Marc Slemko wrote:

> On Fri, 14 Feb 1997, Dean Gaudet wrote:
> > I put Roy's l_c perf patch on HotWired's servers for a few hours... but
> > the FIN_WAIT_2s flew through the roof, so I had to go back to
> > -DNO_LINGCLOSE.  (IRIX 5.3).  There was some speculation that this might
> > help the FIN_WAIT_2 situation, but it doesn't look like it.  As far as
> > performance goes I didn't really get a chance to compare, I'll try again.
> Most important is the performance comparisons between the old
> lingering_close and the one with Roy's patch, although I guess the
> difference between no l_c and the new l_c tells other things that are more
> important in the long-term analysis.
> > 
> > One idle thought I had was that we might try playing with the order of
> > things -- do the shutdown(sd,1) after a select() timeout or a successful
> > read(). 
> Umm... I'm not sure that is nice.  The half close is the only way for the
> client to know that we are terminating the connection.  If we don't do it
> until later, the client won't know that the connection has been closed.
> If the client doesn't know the connection is closed, it can't close its
> half.  That would extend the lifetime of _all_ connections beyond what it 
> is right now even with lingering_close.
> > 
> > The other thing I found, just like Chuck said a while back, I played with
> > the Min/MaxSpareServers settings and removed all the huge spikes from my
> > load averages.  The change in load *could* be related to locality of
> > reference in the L2 cache, but I'm not convinced that is the only thing.
> > We're talking a difference of 50% vs 20% "cpu usage" as reported by the
> > status module.  (I'd really like to revamp status to do 5 minute rolling
> > averages.)
> Played in what way?  Just increased them both?

View raw message