SUMMARY(2): vmunix: malloc_mem_alloc: no space in map

From: Kjell Andresen <kjell.andresen_at_usit.uio.no>
Date: Fri, 20 Nov 1998 22:49:05 +0100 (MET)

Q: http://www.uio.no/~kjell/du/malloc.txt
A: http://www.uio.no/~kjell/du/malloc-sum.txt

Kjell Andresen Systems administrator, University of Oslo, Norway
                Center for Information Technology Services and
                Department of Geophysics
Address http://www.uio.no/~kjell/eget/adresser.html

---------- Forwarded message ----------
Date: Fri, 20 Nov 1998 13:38:24 -0800 (PST) = 22:39 MET
From: Steve VanDevender <stevev_at_hexadecimal.uoregon.edu>
To: Kjell Andresen <kjell.andresen_at_usit.uio.no>
Subject: Re: SUMMARY: vmunix: malloc_mem_alloc: no space in map

At 22:20 MET Kjell Andresen writes:
>
> We've gone through a few patches provided by a DEC support
> representative for the "malloc_mem_alloc: no space in map"
> problem while still having NFS hangs or system crashes.
>
> Including DU v4.0D patch #3 (internal Compaq released tonight afaik)?

Our systems are currently at 4.0D patch #2. The NFS server that
was having problems is currently running with an additional patch
DEC support calls "moiz_fix_2", although the patch alone (without
the tuning change) did not seem to solve the problem.

> However, we've managed to keep our system running for almost a
> month now without problems after setting "vm:
> vm_ubcseqstartpercent = 30" in /etc/sysconfigtab (and leaving
> vm_ubcseqpercent at its default of 10).
>
> We have the latest patch from DEC engineering, but haven't
> rebooted with it yet because we haven't needed to.
>
> Tnx! Just a little question: Wht or which tools did you use in order to
> check system resources and track the right parameter for kerneltuning?
> Did you have to build a new kernwl and then reboot (is front of the month
> uptime)?

I used vmubc to watch the system and noticed that NFS activity
involving very large files would tend to flush out nearly all the
UBC pages and tended to precede NFS hangs or system crashes.

The critical suggestion of reducing vm-ubcseqstartpercent from
the default of 50 (percent of memory) to 30 came from DEC's 4.0D
"System Configuration and Tuning" manual section 4.7.18,
"Controlling Large File Caching". The manual is available online
at the URL:

http://www.unix.digital.com/faqs/publications/base_doc/DOCUMENTATION/V40D_HTML/AQ0R3FTE/TITLE.HTM

You don't need to build a new kernel to implement this change;
just edit /etc/sysconfigtab and reboot.

We currently have the following vm tuning changes in our
sysconfigtab.

vm:
        vm-ubcseqstartpercent = 30
        vm-mapentries = 16384
        vm-vpagemax = 65536

Feel free to forward this back onto alpha-osf-managers. I
actually asked pretty much the same question as you did and
posted a summary a while back based on the information we had
then, but haven't gotten to sending this information to the list
yet.
Received on Fri Nov 20 1998 - 21:50:13 NZDT

This archive was generated by hypermail 2.4.0 : Wed Nov 08 2023 - 11:53:38 NZDT