Latest FC kernel still have SMP bugs?

[Nigel Wade]

WipeOut wrote:
> Nigel Wade wrote:
> 
>> Mitch Wiedemann wrote:
>>
>>> Is the FC1 2.4.22-1.2174.nptl kernel still having problems with SMP 
>>> and hanging?
>>> I'm due to install some MP servers soon and want to avoid problems 
>>> (imagine that!).
>>>
>>> If said kernel is still having trouble, what's the easiest fix?
>>>
>>> Thanks in advance.
>>>
>>
>> I'm currently investigating SMP here. When I tried stock FC1 and the 
>> latest 2.4.22-1.2174.nptl I had severe problems with LDAP. I didn't 
>> get any hard lockup, but I would have delays of between 4 and 300 secs 
>> in LDAP response times. I have a strong suspicion that the delays are 
>> related to FUTEX, but that's hard to prove.
>>
>> I found that disabling hyperthreading is the answer to my problems. 
>> When I disabled it in the BIOS LDAP started behaving itself.
>>
>> I'm currently stress testing the LDAP server and it's managing ~40,000 
>> requests per minute with the peak search times of around 5ms. I read 
>> in some other threads that mounting/unmounting NFS and Samba 
>> filesystems lead to hard lockups so I'm mounting/listing/unmounting 
>> one of each every half hour. The test will be left running for the 
>> next 72 hours to see what happens...
>>
>>
> Compile a SMP kernel from vanilla kernel.org source and it works like a 
> dream.. Even with HT enabled.. :)
> 
> Later..
> 
> 
> 
> 

My system crashed at 23:35. The cron job to mount/unmount filesystems was 
set to run at :10 and :35 every hour. Given what other people report in this 
and other threads I'd say either mount or unmount was the culprit.

I think I'll be building my own kernel now.

-- 
Nigel Wade, System Administrator, Space Plasma Physics Group,
             University of Leicester, Leicester, LE1 7RH, UK
E-mail :    nmw at ion.le.ac.uk
Phone :     +44 (0)116 2523548, Fax : +44 (0)116 2523555