[Mageia-sysadm] multiple builds on each nodes

[Thomas Backlund]

nicolas vigier skrev 11.4.2011 18:57:

>>
>> I removed jonund from buildsystem to run some tests on it. I am
>> currently rebuilding kernel and kernel-tmb at the same time on it.
>>
>> And it seems the CPU does not spend a lot of time waiting for IO :
>> Cpu(s): 94.2%us,  5.8%sy,  0.0%ni,  0.0%id,  0.0%wa,  0.0%hi,  0.0%si, 0.0%st
>> Mem:   8181404k total,  7710776k used,   470628k free,   427864k buffers
>> Swap:  8185076k total,    85332k used,  8099744k free,  4996808k cached
>>

Yeah, sorry, I wasn't clear enough... as seen below:

>> When this is finished, I will rebuild kernel and kernel-tmb separatly
>> to compare time.
>
> So the logs of build of kernel and kernel-tmb at the same time :
> Mon Apr 11 15:26:44 CEST 2011 : startbuild for kernel-2.6.38.2-4.mga1
> Mon Apr 11 15:27:01 CEST 2011 : startbuild for kernel-tmb-2.6.38.2-1.mga1
> Mon Apr 11 16:17:26 CEST 2011 : endbuild for kernel-tmb-2.6.38.2-1.mga1
> Mon Apr 11 17:43:47 CEST 2011 : endbuild for kernel-2.6.38.2-4.mga1
>
> It took 50 minutes to rebuild kernel-tmb (with kernel building at the
> same time).
>
> And 2 hours and 17 minutes to rebuild kernel (with kernel-tmb building
> during the first 50 minutes).
>
> It seems most of the time to rebuild kernel package was spent to install
> files and generate all binary packages.
>

Yep, it takes 2 kernel builds to get all cpus maxed out on jonund


and here is where the io comes in:

during unpackaging and repackaging of big tarballs, and when compressing 
the modules and generating the binary rpms (especially kernel -debug 
rpms are huge)... at this point cpu goes down and io is kept busy...

So here it would help if the other build would be running on another disk...