Skip site navigation (1)Skip section navigation (2)
Date:      Tue, 19 Oct 2004 13:53:36 -0400
From:      Ken Smith <kensmith@cse.Buffalo.EDU>
To:        freebsd-hubs@freebsd.org
Subject:   Re: Just curious - cvsup mirror connect rate?
Message-ID:  <20041019175336.GA26580@electra.cse.Buffalo.EDU>
In-Reply-To: <20041019170412.GV42886@hex.databits.net>
References:  <20041018203143.GB23076@electra.cse.Buffalo.EDU> <20041019170412.GV42886@hex.databits.net>

next in thread | previous in thread | raw e-mail | index | archive | help
On Tue, Oct 19, 2004 at 12:04:12PM -0500, Will Andrews wrote:

> At last count, this server had served 159,208 connections since
> it was last restarted July 13th & rarely sees load avgs above 2.0
> & almost never above 10.0.  On a particularly good day, it serves
> 1500-1600 clients.

Thanks.  I was typing faster than I was thinking yesterday - looking
at the wrong number.  The number I was looking at was the total connect
requests, which at the time was including tons of rejections since it
was pegged at max connects and stayed there...

I've seen this sort of thing happen before, this is why I put one of my
servers in place initially and then pass on the name to another volunteer
site.  The initial load when a 'cvsupX.freebsd.org' server starts to
respond again after a long time of being down is kinda scary and I try
not to freak out the new volunteer site...  Basically all the end-user
types who picked a cvsup server, stuck it in their nightly cron job,
and then forgot about it also forget to check their log files and don't
notice they're not getting updates any more.  So when the server does
finally come back online a large percentage of the machines it had
been servicing before are still trying to get their updates from it.
But now they're all severely backlogged and stay connected for quite a
while during their first successful connect.  Same seems to apply to some
sites that set up their own internal mirror - some don't seem to notice
when their upstream feed stops working...

I've just never seen it this bad before.  Usually the initial shock
phase ends anywhere from 4 to 6 hours after I make the DNS change.
This time it took over 24 hours and the machine was saturated the
entire time (I upped the connect limit from 30 to 40 part way through
because the machine seemed to be handling it OK and even there it
remained saturated).  The load average spent most of its time above
20, not dropping into the 10's until close to the end...

Now it's back down to less than 1 with five to ten cvsupd's running
at a time...  So now I can give it to someone else... :-)

-- 
						Ken Smith
- From there to here, from here to      |       kensmith@cse.buffalo.edu
  there, funny things are everywhere.   |
                      - Theodore Geisel |



Want to link to this message? Use this URL: <https://mail-archive.FreeBSD.org/cgi/mid.cgi?20041019175336.GA26580>