On 05/11/05 18:28, Chris wrote:
On 5/11/05, Marian Hettwer <[EMAIL PROTECTED]> wrote:
On Mi, 11.05.2005, 14:34, Steven Jurczyk sagte:
**Using SCHED_ULE on SMP/HT machine broke applications which use
libpthread/libthr, probably at context switching between threads... The
applications simply hang and they aren't killable (kill -9 <pid> don't
work).. Usually this also break kernel shutdown (can't flush some inodes
or blocks)...

right! I have the same problem. My Kernel is basicly a GENERIC with SMP and ULE added (4BSD removed).

([EMAIL PROTECTED] <~>)$ uname -a
FreeBSD siteop-8.mobile.rz 5.4-STABLE FreeBSD 5.4-STABLE #7: Wed May 11
12:03:35 CEST 2005
[EMAIL PROTECTED]:/usr/src/sys/i386/compile/SITEOP-8  i386


This is very big problem for people who want use mysql_server on SMP
machines...

indeed. The MySQL server won't start and is not killable. I had to switch back to SCHED_4BSD :-/

I would be available for testing packages, as this machine is not in
production.

it's a dual xeon 2,8 ... you can find detailed information about it at:
http://unixoid.de/freebsd/dmesg.xeon

:)

Seems the 4+ cpu crash fix might have broken it at a guess since RC4 onwards it broke, I am interested in how this goes as I was planning on upgrading a SMP box to 5.4 and ULE this month.

Have you guys tried this without SMP?

From the sched_ule(4) man page:

"BUGS
As an experimental scheduler, sched_ule is not enabled by default due to a number of known issues, including weak performance with several known workloads, and reports of instability. Deployment of sched_ule in production environments should be done cautiously."

The less well-known 5.4 Open Issues page (http://www.freebsd.org/releases/5.4R/todo.html) speaks to this as well:

"Many improvements have been made to the ULE scheduler in 6-CURRENT. These should be merged back to 5.4. The merging is done but ULE is still known to cause panics for some people, especially on SMP systems. Try it with extreme caution."

In other words, ULE is known to not be stable and will remain so until someone fixes it.

--
Jonathan Noack | [EMAIL PROTECTED] | OpenPGP: 0x991D8195

Attachment: signature.asc
Description: OpenPGP digital signature



Reply via email to