Subject: CATS 1.3F problems (in lieu of -current working)
To: None <port-arm32@netbsd.org>
From: Dr. Stephen Borrill <sborrill@xemplar.co.uk>
List: port-arm32
Date: 11/16/1998 10:46:41
In lieu of getting a working -current on the CATS, I'm sticking with the
1.3F tree provided. However, there are certain (known?) problems,

There are frequent crashes when system system resources are low. What do
I mean by system resources? Don't know; probably a combination of L1
page tables, mbufs, etc. I've installed one of these machines at a site
with an ISDN router over which they often get a lot of mail delivered.
One user joined a mailing list and (seemingly) requested a copy of all
previous mails (they got >1000 in a day). This forked large numbers of
sendmails which seemed to just hang when it got to the DATA stage of the
SMTP transfer. We got the mail backlog cleared but this morning, the ISP
is sending all the mail that's been waiting all weekend for the ISDN
line to come up. When the CATS gets to about 130 processes (with
PMAP_STATIC_L1S set to 192), the machine falls over in various ways. You
get "fork: cannot allocate memory" errors, sometimes it just crashes.
This is reminiscent of various posts I made over the summer detailing
how PMAP_STATIC_L1S seemed to work as expected on a RiscPC, but was
flaky on a CATS. Given the amount of memory in the CATS, I'd like to
ramp up the L1 page table number to something much higher to give it a
fighting chance of being a useful server, but there's no point if it
can't cope with more than about 130 processes.

The hanging sendmail processes may or may not be related to a problem I
reported before (but didn't get any response to). With my CATS machine
here (being used as a mail/IMAP server amongst other purposes), I find
that after about 48 hours, it is often unable to successfully open a TCP
connection to the outside world. For instance, sendmail just reports
read errors when trying to deliver mail, ftp reports "421 Service not
available", etc. A reboot _always_ cures this and so I've got it
rebooting every night at midnight, just to be safe. Bang go the long
uptimes I was so proud of.

-- 
Dr. Stephen Borrill, Network Computer Technical Manager
Xemplar Education Ltd                         Tel: +44 (0) 1223 724 267
The Quorum, Barnwell Road                     Fax: +44 (0) 1223 724 324
Cambridge, CB5 8RE, United Kingdom            WWW: http://www.xemplar.co.uk/