[csw-maintainers] [csw-buildfarm] BO Buildfarm available again

[Dagobert Michelsen]

Hi James,

Am 15.01.2010 um 13:03 schrieb James Lee:
> On 14/01/10, 21:43:09, Dagobert Michelsen <dam at opencsw.org> wrote  
> regarding
>> Is this a problem for you?  Are you really using that many parallel
>> threads?
>
> Yes, but only at times.  OpenOffice.org is the worst and uses all
> threads for long periods (when the disc isn't dying).  Looking at my
> current terminal output tiff says "build8s --> 20 jobs" several times
> but not for long as the whole build only takes 6 mins for everything.
>
> Problem is the wrong word.  Only recently has the build farm had  
> enough
> disc space

I can now easily add more space as I didn't assign everything that
has been installed. Just let me know if you plan big projects ;-)

> and I've built OOo on my own Pentium 2 333MHz, it took 4
> days not 6 hours but did get there.  So it's not a problem but more
> is better, 5 minutes is better than 6 minutes and it won't be too fast
> until it's quicker than pressing a key.

I see. For now I can move the CPUs from the OSOL LDom to the Solaris
LDom until it is more heavily used.

>>> That is on the misleading
>>> side of truthful as there is no dynamic sharing of CPU and memory.
>>> Where load is sporadic as for build hard partitioning is a waste  
>>> of a
>>> good machine.
>
>> I had the impression that the T5220 wasn't used that much at all.  
>> Most
>> of the time the load is beneath 3 or 4
>
> Exactly the problem I mean by sporadic load.  Something like a busy  
> email
> server will have a continual somewhat random stream of messages and  
> will
> have a meaningful average load.  A build the machine is idle until  
> it's
> used, then it's busy.  The load average means little, being low  
> means no
> one is working.  You should only take the load average of when it's  
> busy
> which sounds silly but is right because the load when no one is  
> using the
> machine doesn't matter.  This is where average isn't important, as  
> with
> the average number of legs on people, or rating the fire brigade by  
> the
> percentage of time they are actually putting out fires.
>
> If it's a natural consequence of using LDoms then so be it.  It's a
> shame they can't dynamically share threads, as with unrestricted  
> zones,
> because for very uneven loads it's what is needed.

I can move CPUs around on request if that helps. If it really shows
to be a problem I will add another machine, but first I'd like to try
it with this configuration.


Best regards

   -- Dago