Just curious - cvsup mirror connect rate?

[Will Andrews]

On Mon, Oct 18, 2004 at 04:31:43PM -0400, Ken Smith wrote:
> I'm just curious what people typically see for connect rates to the
> main cvsup mirrors.
> 
> I've got one server here that's desperately in need of more memory
> but it typically services around 2500 connections a day.  Another
> cvsup server I've got access to at another site services roughly
> the same number per day.
> 
> The reason I'm asking is I'm cleaning up some servers that vaporized
> on us.  I've started to take over a dead server's name on a large
> server here for at least a day before I turn it over to a new volunteer
> site because of the inevitable slamming that happens when a new server
> takes over for one that had been down for a while.  I've done this
> a couple times now and despite the initial load once things settle
> down a bit I've been seeing the same typical number - on average
> around 2500 connects a day.
> 
> Until now...  I just took over cvsup5 for a bit because it had been
> down.  In less than 6 hours it's serviced about 11,000 connects.  And
> I'm not seeing tons of "Tree comp failure..." messages, the huge majority
> are reporting success.  It's been pegged at the 30 connect limit ever
> since I made the DNS change almost 6 hours ago...
> 
> So, just kind of wondering what people normally see as average traffic
> for a day.  If cvsupd has been running all the time your machine has
> been up simply taking the current connect number it's allocating and
> dividing by the number of days it's been up is close enough. :-)
> I'm kind of wondering if the number of hits you get is a function
> of what cvsupN number you get - when people first set themselves
> up they start at cvsup1 and go up until they find one that doesn't
> say its full and then set up their cron job to use that one... :-)

I think the imbalance is because people (falsely) rely on a
program called "fastest_cvsup" to choose the mirror to use.  This
program only measures network performance of the cvsup mirrors in
question & thus leads them to use a mirror that is not
necessarily their best choice.  The fact is that the network
performance means little when it comes to cvsup, especially for
updates.  CPU, RAM, disk I/O of the server mean a lot more.

This particularly affects cvsup12 since its outbound link is
frequently saturated.  The statistics are, for this month:

will at sanmateo% for i in `jot -w '%02d' 18 1`;do echo -n "2004/10/${i}: ";grep ^2004.10.${i} cvsupd.log | grep ": +"|wc -l;done
2004/10/01:     1255
2004/10/02:     1218
2004/10/03:     1152
2004/10/04:     1305
2004/10/05:     1349
2004/10/06:     1250
2004/10/07:     1314
2004/10/08:     1253
2004/10/09:     1252
2004/10/10:     1188
2004/10/11:     1212
2004/10/12:      866
2004/10/13:     1265
2004/10/14:     1246
2004/10/15:     1288
2004/10/16:     1267
2004/10/17:     1225
2004/10/18:     1324

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.

Regards,
-- 
wca
-------------- next part --------------
A non-text attachment was scrubbed...
Name: not available
Type: application/pgp-signature
Size: 187 bytes
Desc: not available
Url : http://lists.freebsd.org/pipermail/freebsd-hubs/attachments/20041019/70b8914b/attachment.bin