sendmail performance question

[David A. Sinck]

\_ SMTP quoth Craig White on 4/8/2002 10:38 as having spake thusly:
\_
\_ On Fri, 2002-04-05 at 13:43, David A. Sinck wrote:
\_ > 
\_ > I could stand a few pointers here, as I've reached the bottom of my
\_ > idea barrel and I've got splinters from scraping the bottom.
\_ > 
\_ > Synopsis:  
\_ > 
\_ > sendmail on a dual p3/733 w/ 1G + scsi is 30% of the speed of a p3/667
\_ > w/ 512M + ide.
\_ > 
\_ > That just doesn't seem right; I want to know why.
\_ > 
\_ > If you're not interested, this would be a good time for the D key :-)
\_ > 
\_ > 
\_ > Details:
\_ > 
\_ > Common:
\_ > * same perl version
\_ > * same perl Net::SMTP version
\_ > * same sendmail + sendmail configs (md5sum the same)
\_ > * same input (1k test messages)
\_ > * same network (xxx.3 + xxx.4)
\_ > * same nics
\_ > * both 'mostly idle'
\_ > 
\_ > Slow:
\_ > * dual p3/733, 1G, scsi
\_ > * 5+m return
\_ > * kernel 2.4.9-13smp  (redhat)
\_ >  
\_ > Fast:
\_ > * p3/667, .5G, ide
\_ > * 1+m return
\_ > * local dns server
\_ > * kernel 2.4.9-12 (redhat)
\_ > 
\_ > 
\_ > The slow box uses the dns server on the fast box...and peculiarly, dig
\_ > against it comes back faster reliably based on hand tests when the dns
\_ > server is accessed remotely.  (1-2ms vs 3ms).
\_ > 
\_ > hdparm: buffer cache faster on slow box (scsi) (2x) but slower on
\_ > buffered disk reads (1/2) vs the fast box.
\_ > 
\_ > I reluectantly downed the local firewall on the slow box (albeit still
\_ > behind a router/firewall, I'm not completely insane :-), but that
\_ > didn't change the performance.
\_ > 
\_ > It *appears* as if the box with the local bind on it is forking
\_ > sendmail 3-4x what the non-binded box is, which would account easily
\_ > for the performance....  But if the configs are the same, how is that
\_ > possible? 
\_ > 
\_ > Ideas?  Preferably good ones?
\_ -----
\_ OK - speed issue is vague.

1000 messages leave the app in 1+ minutes on the fast server, in 5+
from the slow server.  I can get real times if you don't want the
vague "1+", but I figured it was immmaterial :-P.

\_ If you are talking about a localhost send, then perhaps it is the time
\_ with which to resolve the sending host - check the file /etc/hosts to
\_ make sure that 127.0.0.1 localhost.localdomain localhost is included and
\_ that the hostname resolves internally.

All test emails are going to a single domain, not hosted on the box;
afaict, there aren't any dns issues with the box.

\_ If you truly believed that it was merely a matter of having bind locally
\_ on the machine, it wouldn't take long 

But I don't.  It's like sendmail's not forking enough or something.

\_ I am curious that the sendmail config files are the same - how could
\_ both machines be willing to accept the mail for the domain in the same
\_ manner? 

1) The target domain of emails is different than the local box, for
   both boxes.

2) Think upgrade.  The old fast box wasn't "big enough" for everything
   it was doing, so the mail responsibilities got migrated to a new
   "better" box.  Hence, the names + configs got shifted from one box
   to the other...but the old box is still up and chugging away.


David
ps: anyone else have problems with English quoting rules vs
your-programming-language-here quoting rules?  I can ignore it until I
trip across a contraction.  :-)