kernel crash on Intel(R) Xeon(TM) CPU 2.66GHz 4-cpus machine

[Andrea Puglisi]

Hi Jean-Eric,

it's almost four days that two processes (simulations) are running on the 
two processors of my machine, with hypertrading disabled. No more freezes! 
I suspect this implies two possible explanations:

1) two simulations instead of four means also a lower use of memory, 
therefore an eventual physical memory problem does not show up; anyway it 
says that the current memory use is

    Mem:   1034712k total,   910956k used,   123756k free,   184972k buffers
    Swap:  1951888k total,      220k used,  1951668k free,   322360k cached

  (that is: phys. memory is almost entirely used)


2) there is a problem with kernel 2.6.15-26-686 together with hypertrading 
(or a problem with the hypertrading feature of my machine)

Have you any kind of definitive test to discriminate between these two 
hypothesis?

  Ciao

  Andrea


On Mon, 21 Aug 2006, Jean-Eric Cuendet wrote:

>
>>> OK, one thing left to investigate. Just to be sure, how long have you
>>> run the test? It's about 2h for a single step.
>>
>> It took about 30 minutes to perform all tests, then it started it again
>> and I quit it (I thought it was a infinite loop). Should I do it many
>> times?
>
> It's an infinite loop, indeed.
> IMO, if you did it 1 time, it's OK.
> Though at your place, I would have let it run the whole weekend...
>
>>> It's a HyperThreading processor. It's *seen* as 2 processors / CPU but
>>> in fact it's just useless in the majority of situation. We ended up
>>> disabling it in the BIOS for ours...
>>
>> what does it mean exactly? if I disable it, does it mean I lose
>> computational power? 2 processors instead of 4 is of course worst, but
>
> Yes, 2 are worst than 4. But in HT cases, 2 could be better than 4! :-)
>
>> if two of them are only virtual, then the total power (number of
>> operation per clock tick) should remain similar, am I right?
>
> Not exactly. Hyperthreading is pseudo-processors. In certain cases, you
> get 4 processors. But in the majority of cases, they are useless. In
> fact, you get around 30% more processing power with HT enabled on some
> workloads. And 5% less on others...
> If you do certain types of processing with multiple threads, then they
> could help. Disable them at first, then when your freeze is solved,
> re-enable them and bench your app. It could help but it depends on the
> workload.
> HT was marketing hype on Intel side IMO.
>
>> Now I disabled HyperThreading, but I could not disable one cpu (at least
>> on the bios setup). Therefore now my kernel has recognized two
>> processors. I've just launched two simulations. Let's see what happens.
>
> OK, to disable the other CPU, run a not-SMP kernel.
> Ben? Is there a way to disable one CPU on the kernel command line?
> Else, run the i386 kernel (linux-image-386)
>
> Bye.
> -jec
>
> -- 
> Best regards / Salutations.
>
> Jean-Eric Cuendet
> Senior developer / Technical support
> Riskpro Technologies SA
> Av. Louis-Ruchonnet 2
> CH-1003 Lausanne
> Switzerland
>
> Direct    : +41 21 637 0123
> Principal : +41 21 637 0110
> Fax       : +41 21 637 0111
> Skype     : jec.rptec
> Web       : http://www.jesc.ch
> --------------------------------------------------------
>
> -- 
> kernel-team mailing list
> kernel-team at lists.ubuntu.com
> https://lists.ubuntu.com/mailman/listinfo/kernel-team
>
>