We upgraded last week from OSF 3.2c to Digital UNIX 4.0b on our
alpha email server. We are running Berkeley Sendmail-8.8.4 and are
pretty close to capacity.
The system runs sendmail and our pop server; general users are not
allowed to log on interactively. On Monday, our first busy day since
the upgrade, the system clogged up. The sendmails don't seem to
finish. They just accumulate, and the system gradually grinds to
a halt. When we get up to about 4,000 sendmails, we reboot.
The qpopppers seem to come and go without problem. Other hosts
that connect to our sendmail port seem to get hung up after DATA.
Eudora clients that connect timeout when connecting to our SMTP
prot.
We've discovered if we kill the sendmail daemon, things quiet down -
the queue dissipates and the number of sendmail processes running
dies down. The load average at the peak does not look bad, but
we do start running out of memory and start paging/swapping.
Also, when it is not very busy, the system seems to work ok.
Maybe the multithreaded kernel is somehow involved in the problem?
I don't believe we got a huge increase in mail volume since a week ago.
One idea is that somehow /var/spool/mqueue is getting locked and
all the sendmails are waiting, but I have no evidence of that.
We are desperate at this point - this is the email server for
22,000 accounts - does anyone have any ideas?
It has even been suggested that we install DEC sendmail, so we can
bug DEC for support.
Regards, Michele Tomkin
Workstation Software Support Group
UB Berkeley
Received on Thu Apr 03 1997 - 03:57:29 NZST