System capacity

[Hirsh, Joshua]

    [ The following text is in the "iso-8859-1" character set. ]
    [ Your display is set for the "US-ASCII" character set.  ]
    [ Some characters may be displayed incorrectly. ]

 Since we're on the subject, I was running some performance testing on an HP
DL380 the other day (dual 3Ghz with 2Gb of RAM and 2 RAID-1 partitions using
15k RPM drives).

 I was able to push just under 1200 messages per minute (1.7 million per
day) using MS 4.35.11 and Postfix 2.1.5, SA 3, DCC, Razor and RBLS. For my
tests I made the following changes to the system:

 Caching nameserver
 Ran nscd
 Set /var/spool/MailScanner as tmpfs
 Disabled atime record keeping on /var/log
 Disabled syslog synching in /etc/syslog.conf


 I only had a chance to test Sendmail and Postfix at the time. I'll try out
Exim at another date. However, I couldn't get Sendmail to run anywhere near
the performance levels I saw with postfix. I could only get Sendmail to run
with 400 messages per minute with the system configured identically.

 Even without MailScanner, Sendmail performed extremely poorly. As a pure
relay server, Postfix was able to handle close to 2400 messages per minute
(almost 3.5 million per day). Sendmail could only make it up to 900 messages
per minute.

 The system was configured to relay all data to another server running
smtp-sink.

 Email was generated using Postal with random sizes of up to 100k per
message from various source machines.

 With MS running, the load average with Postfix was around 12, however
messages were still being sent with a very minimal delay (one minute at
most). During the Sendmail test, the load was around 15 to 16 and messages
were being delivered with an average delay of 3 to 5 minutes.

 Without MS running, the load average with Postfix was 4. With Sendmail, the
load was around 8 to 12.





 Now of course, take all performance measurements with a grain of salt,
unless you run them for yourself. Every system is different and every
configuration of said system will interoperate differently. These are just
my own figures that I've run across during my own performance testing. I'm
sure Sendmail could have been tweaked to work better, but I ran both MTA's
"out of the box" with default configurations (except for defining a relay
and transport host).

 Real world production is always going to be different from testing like I
did. For one, I'm not Sendmail email from hundreds to thousands of different
source addresses like a real system of this size would expect to receive, so
the caching of RBL lookups only has to remember a handful of hosts during my
tests.



 Cheers,

--
Joshua Hirsh
Systems Administration
Partner Solutions Inc.
PGP/GnuPG ID: 0xD12A3B59

------------------------ MailScanner list ------------------------
To unsubscribe, email jiscmail at jiscmail.ac.uk with the words:
'leave mailscanner' in the body of the email.
Before posting, read the MAQ (http://www.mailscanner.biz/maq/) and
the archives (http://www.jiscmail.ac.uk/lists/mailscanner.html).

Support MailScanner development - buy the book off the website!