Re: suspiciously terrible performance of 2.5.1 beta on OpenBSD/amd64 (now a crash...)
2006-08-03 11:05:31
On 2006-08-02 10:31:43 -0700, Paddy Sreenivasan wrote:
> Can you please clarify what you mean by "nearly-2.5.1". Is it 2.5.1b2
> or 2.5.1b1? There are lots of memory issues fixed using valgrind
> and other tools in 2.5.1b2 release.
>
2.5.1b2 (and CVS HEAD from yesterday morning) exhibit the problem
in driver when holding disk is turned on. Will stick to 2.5.1b2
for bug reporting.
For completeness (I don't quite see how it could cause the problem, but...):
I also had to apply the trivial EOVERFLOW->EINVAL patch I attached to
bug #1530403 (without which one can't build on openbsd).
> If you are using 2.5.1b2, please file a sourceforge bug for the crash
> (https://sourceforge.net/tracker/?func=add&group_id=120&atid=100120)
> Use group v2.5.1b2
>
Was going to, once I do a bit more discovery...
One thing I found out that I haven't firmly established is a problem yet,
but is likely worth a mention:
OpenBSD 3.9's standard kernel apparently still has a
surprisingly low hard upper limit on per-process memory space
(1 GB, even on AMD64...). I expect that could cause problems...
http://marc.theaimsgroup.com/?l=openbsd-misc&m=115320322614641&w=2
Maybe I should try some small chunk sizes and --without-mmap ...
|
|
|