Re: cvs commit: src/sys/kern sched_ule.c (fwd)

Jon Noack wrote:
> Tony Arcieri wrote:
> 
>>On Tue, Dec 14, 2004 at 05:08:43PM -0600, Jon Noack wrote:
>>
>>>I thought about trying this last night when I saw that ULE was
>>>resurrected.  Make sure you also grab kern_sig.c:
>>>http://lists.freebsd.org/pipermail/cvs-src/2004-December/036757.html
>>>
>>>I can't say whether those 3 files are all you need, just that I would
>>>also include kern_sig.c... ;-)
>>
>>Rebuilt with kern_sig.c from -CURRENT, everything seems fine, as far as I
>>can tell.  Are there really any substantial changes in kern_sig.c and
>>kern_switch.c that would affect the stability of 5_STABLE (and does
>>UMA in 5_STABLE ensure thati proc_fini() won't be called?)
> 
> 
> I don't know about kern_switch.c, but the change in kern_sig.c fixes #2 on
> Jeff Roberson's list of bugs in ULE (from a few days ago):
> http://lists.freebsd.org/pipermail/freebsd-current/2004-December/044332.html
> 
> 
>>I'd just contend that in the case of my system, 5_STABLE with the 4BSD
>>scheduler is not stable, or at least the script I'm running is somehow
>>exhausting system resources to the point that the system becomes unusable,
>>and this problem isn't exhibited with the ULE scheduler.  Regardless, the
>>script was causing the 5.3-RELEASE GENERIC kernel to panic, and rendered
>>the system completely inaccessible with a kernel built from the latest (as
>>of about 5 days ago) RELENG_5 kernel with the 4BSD scheduler.
>>
>>So, I'd be very grateful if ULE could be merged into RELENG_5 as it would
>>dramatically improve the stability of at least my server.  Has anyone else
>>with a dual amd64 system had problems like this post 5.3-RELEASE?  I know
>>crashes under heavy MySQL load on dual amd64 systems were a problem
>>before, but I thought that had been resolved.
> 
> 
> I think removing the #error and putting a note on boot (and in UPDATING)
> that it may still be unstable is a good idea.  However, Scott Long has
> expressed reservations
> (http://lists.freebsd.org/pipermail/freebsd-current/2004-December/044341.html)
> and his opinion counts orders of magnitude more than mine.
> 
> Jon
> 

I'm definitely not against these fixes going into RELENG_5, but I would 
like to see some significant testing be applied to them in HEAD first,
especially to changes that are not confined to just sched_ule.c (and
sched_4bsd.c).

Scott