increasing priority of sendmail? or work around

[Mike_Ca Li]

About 2500 email messages are sent from my legacy box daily using sendmail. The box is overloaded and management put off getting a more powerful box.
Mail messages would accumulate in /var/spool/mqueue for hours and mailq would display about 60-90 queued at any moment in time. The mail message size varies between a few 100K up to 3MB. I tried to do sendmail -q0 and I also used renice -40 sendmail.pid but that does not help much.
Any workaround or can I move the files in /var/spool/mqueue to another box with sendmail? Thanks.

[RAC_1]

Is this server used only for mailing?? If yes, no need to do renice on sendmail pid. Else you may look at WLM (work load manager) and PRM (process resource manager) products. Both are licensed. So additional $$ required. May be you can add some more RAM and check.

Anil

[Sanjay_6]

Hi Mike,

Maybe this link will help,

http://www2.itrc.hp.com/service/cki/docDisplay.do?docLocale=en_US&docId=200000066259197

the itrc doc id is 4000048583.

Hope this helps.

regds

[Patrick Wallek]

What type of box is this? What OS is it running? Unless the machine is very busy doing other things, 2500 messages really should not be that difficult to do. Over a 24 hours span (24 hours = 1,440 minutes) that's slightly less than 2 messages per minute. That should be doable.

Are there a lot of messages that get generated at once? What is the real time span for the 2500 messages to be generated?

[Mike_Ca Li]

box is not only used for mailing. Heavily used for reporting, searches and transaction entries. < 5% idle between 9am-6pm. Span of use is mostly between 7am-11pm, when there are most of the 2500 emails generated. HPUX 11.11. Sendmail 8.9.3

Thanks.

[Shannon Petry · Accepted solution]

Sendmail is not very often CPU bound, so re-nicing the sendmail process will not do anything.

Sendmail is I/O bound more commonly, and this can be Network, Disk, or both.


vmstat and iostat should give good indicators if the system is disk bound. Add another controller and couple disks, stripe together mail spooler directory.

If the problem does not appear to be disk, check network I/O. Add a designated NIC for internet traffic, especially email.

Regards,
Shannon

[Steven E. Protter]

The box is overloaded.

so sendmail may not be getting enough cpu.

the problem with giving it enough cpu with renice (nice means higher nice, easier on system, lower priority) is that these important reports and other activity will be effected.

I think renice isn't going to help much. Overloaded is overloaded.

You could take a shot at tuning the box in general. You might be able to get some balance.

http://www1.itrc.hp.com/service/cki/search.do?category=c0&docType=Security&docType=Patch&docType=EngineerNotes&docType=BugReports&docType=Hardware&docType=ReferenceMaterials&docType=ThirdParty&searchString=UPERFKBAN00000726&search.y=8&search.x=28&mode=id&admit=-682735245+1102026370706+28353475&searchCrit=allwords

Management made a mistake putting off the upgrade. Use this data to convince them to change their minds.

If the box is upgradeable, maybe some more memory or another CPU would help. Because sendmail is not CPU bound, I kind of doubt it.

You might be able to manipulate the sendmail.cf file and allow more simultaneous mail sessions in the hope that in brief moments of lower system use a whole lot of mail gets squirted through.

I'd also collect some performance data and see where the bottleneck is. If its network, you're done, stick a fork in it because there is nothing you can do short of splitting the functionality from this server.

If its something else, you might be able to remedy it.

Scriptset attached, but modification WILL be required to run them on 10.20.

SEP

[Mike_Ca Li]

Thanks everyone for all the inputs. I'll do loggings and present to mgmt for more Memory or CPU.

[Bill Hassell]

Running all of these different processes on the same box is a recipe for disaster. You'd be a lot better off getting another server dedicated just to the sendmail task. Of course, if the mail is generated by the processes on this box, you may have not only a disk overload but a LAN overload. sendmail can't send anything faster than the remote end is willing to accept the messages. Your mail bottleneck may some overloaded PC running Exchange with a ton of spam filters. I have seen a 50Mhz send 2000 messages per hour with one processor and not breathe hard at all. But the destinations were all HP-UX boxes, not PC servers.

As far as the mqueue directory, using something like NFS to export the files will probably slow things down even more. NFS is intolerant of network overloads. Perhaps setting up a separate subnet (and add a new LAN card) just for email will help.