Load Problem
Julian Field
mailscanner at ecs.soton.ac.uk
Fri May 28 20:32:13 IST 2004
Can someone add that reply to the FAQ please? In my view it's a very good
summary of troubleshooting load problems on busy mail servers, and contains
very many useful hints for people.
At 20:18 28/05/2004, you wrote:
>Ricardo Bernardes wrote:
>
>>Hello,
>>
>>i'm using RH8, ClamAv and MS on a PIV 2,8 box with 512 Mb of Ram.
>>
>>Whenever a user sends an heavy mail (e.g. about 6Mb) to all other
>>users (we're about 130) my server takes a lot of time to go back to
>>normal.
>>the POP3 connections get very slow and some even time out; the SSH
>>network (over LAN) connections times out
>>after everyone downloads the e-mail, it goes back to normal
>>
>>any thoughts how to overpass this situation?
>>
>>
>>thanks
>>ricardo
>
>
>Once again, basic troubleshooting applies.
>
>It sounds like an email server issue rather than MailScanner issue. Is
>your POP server is the same box as the mailscanner/clamav box?
>
>Is the sender sending a single message to many recipients, or sending
>many (similar but separate) messages to each user?
>
>If the former, what is your back-end email system? If it's Cyrus
>(highly recommended), enable single-instance message store. If it's
>not, consider switching to it using Simon Matter's RPM's. By doing so,
>if a user sends a message to N recipients, the data of the message gets
>written once and for all other recipients, a hard link is made. Note
>this means you don't write N*(message-size) files. Also note you'll
>have to tweak your Cyrusv2 mailer options to accept more than the
>default of 10 or so recipients per message.
>
>The idea here is that if you use hard links, the message data is
>identical for all users and refers to the same disk blocks and thus will
>remain cached in RAM while your users download their copies. Unlinking
>the hard link (inode) is also much quicker than freeing all the disk
>blocks plus the inode.
>
>To help spread out the load, have your users poll for new mail no more
>frequently than 5 minutes, preferably 10. If they are impatient and
>expecting new mail from someone, let them click on "get messages" manually.
>
>When your user sends the mass mailing, run "vmstat 1" and carefully
>study the output (are you swapping? If so, get more RAM.) Many pages
>in/out? Get more/faster disks and stripe them. Use "iostat -x 1" to
>also get more detailed information on disk utilization. Could you
>improve your filesystem performance? E.g., mount the filesystem with
>-noatime. Switch filesystem types (e.g., ext2 -> some journaled
>filesystem, such as Reiserfs [mount with -notail], which is very good
>with small files). Is your CPU idle time zero? Get more/faster CPUs
>(though 2.8 GHz P4 ought to be plenty for 120 people). If you are
>running out of CPU, look at the output of "top" to see what else might
>be contending with it. Consider running POP service at a higher "nice"
>priority, or MailScanner + your MTA at a lower "nice" priority.
>
>Is your network saturated? The world should be all 100Mbit full duplex
>switched, but you never know. Try running "gkrellm" and watch your
>network traffic volume (note it reports KB/sec, not Kb/sec, so on 100Mb,
>expect 8-11 MB/sec max). If you're saturating 100Mbit (highly doubt it
>unless all the users are downloading in the same minute), move to
>Gigabit on your server and make sure the traffic intelligently fans out
>to switched ports to your users.
>
>Finally, you have the option to Punt: don't send large messages! (Put
>the attachment on a file or web server and send the users a link.)
>
>Hope this helps,
>Chris
>
>-------------------------- MailScanner list ----------------------
>To leave, send leave mailscanner to jiscmail at jiscmail.ac.uk
>Before posting, please see the Most Asked Questions at
>http://www.mailscanner.biz/maq/ and the archives at
>http://www.jiscmail.ac.uk/lists/mailscanner.html
--
Julian Field
www.MailScanner.info
Professional Support Services at www.MailScanner.biz
MailScanner thanks transtec Computers for their support
PGP footprint: EE81 D763 3DB0 0BFD E1DC 7222 11F6 5947 1415 B654
-------------------------- MailScanner list ----------------------
To leave, send leave mailscanner to jiscmail at jiscmail.ac.uk
Before posting, please see the Most Asked Questions at
http://www.mailscanner.biz/maq/ and the archives at
http://www.jiscmail.ac.uk/lists/mailscanner.html
More information about the MailScanner
mailing list