more on bogged down server

[Rick Stevens]

On Wed, 2006-04-12 at 13:13 -0700, Harold Hallikainen wrote:
> > On Wed, April 12, 2006 12:37 pm, Rick Stevens said:
> >> On Tue, 2006-04-11 at 08:20 -0700, Harold Hallikainen wrote:
> >>> >
> >>> >> On Mon, 2006-04-10 at 15:53 -0700, Harold Hallikainen wrote:
> >>> >>> >
> >>> I REALLY appreciate all the help on this list!
> >
> > Me too. Even when I haven't asked for it and something comes across . . .
> >
> >>
> >> Check with your ISP to see if they changed the polling intervals or
> >> any other parameter having to do with your transmission pipe.  Sometimes
> >> they add a lot of new clients onto your ring, so they shorten your
> >> poll time to accommodate the new users.  My ISP did that to me on my
> >> cable modem and I raised holy hell with them.  My poll period was down
> >> to 5-10mS!  Ridiculous!  I told them I wasn't paying $40 a month for
> >> farking 9600-baud dialup speeds.
> >
> > Okay, so how do I tell if that's my problem? My ISP changed our DSL link
> > to
> > a different piece of hardware and the speed went from 6896 to 640k, the
> > Qwest default. They've fixed that, but it still appears that my downloads,
> > though much faster, are still not what they were before the move, even
> > though the older equipment only allowed me to train at 6896 instead of
> > 7168,
> > which is what I'm trained at now.
> >
> > Karl
> 
> 
> Still learning how all this stuff works (thanks especially to the list).
> Communications speeds still seem ok (my DSL is 6M down and something less
> up). My server just seems to be bogging down. If communications were slow,
> I guess a lot of httpd processes would start to slowly send the data out,
> or is there a buffer somewhere that can handle that more efficiently? If
> we were I/O bound, it doesn't seem like that'd result in a large cpu load.
> 
> Looking at top, even if there is just one instance of httpd, it will go to
> 100% CPU, or very close to that. I'm assuming it's SUPPOSED to do that,
> just not for very long. When there are lots of instances of httpd, the
> %CPU in top for each drops, but they add up to near 100%, and the total
> %cpu is close to 100%. But, I guess that's ok. If the load were exactly
> 100%, the load average would show up as 1.00, right?
> 
> Now, it's running about 20. sendmail stopped accepting connections at 12.
> As mentioned yesterday, I've added robots.txt and told the search engines
> to not search the directory with the huge pdfs (which is where I'm
> thinking most of the traffic is coming from). I've also put the
> crawl-delay in robots.txt at 60 seconds to avoid those once a second
> requests. But, stuff is still piling up (they may not have read robots.txt
> yet).
> 
> I'm running version 2.0.something of httpd that ships with FC4. I see
> there's now version 2.2 available. It's supposed to handle large files
> better, among other things. I guess I'll give that a try. Others have
> suggested more config file changes (getting rid of mod-perl, etc.) to make
> httpd more efficient.
> 
> I'll keep working on it. Meanwhile, off to restart httpd so I can get mail
> again...

Have you done the "vmstat 3" thing yet to see if you have context
switching going nuts?

----------------------------------------------------------------------
- Rick Stevens, Senior Systems Engineer     rstevens at vitalstream.com -
- VitalStream, Inc.                       http://www.vitalstream.com -
-                                                                    -
-    If Windows isn't a virus, then it sure as hell is a carrier!    -
----------------------------------------------------------------------