Re: Architecture qualification
Le 15/05/2012 22:27, Michael Banck a écrit :
> On Tue, May 15, 2012 at 09:45:43PM +0200, Julien Cristau wrote:
>> On Tue, May 15, 2012 at 20:42:14 +0100, Adam D. Barratt wrote:
>>> On Tue, 2012-05-15 at 20:42 +0200, Julien Cristau wrote:
>>>> On Tue, May 15, 2012 at 16:18:19 +0100, Adam D. Barratt wrote:
>>>>> In an effort to stop this stalling any further / longer, I propose
>>>>> sending [1] to each of the port lists, probably some time tomorrow.
>>>>>
>>>>> Comments / changes / updates / whatever welcome.
>>>>>
>>>> I'd add a concern about the mips buildds to the arch qual page (not sure
>>>> how to phrase it).
>>>
>>> Assuming it's the stability issue, then maybe re-using weasel's
>>> "unstable under load" which we used for the porter box?
>>>
>> There's that, and there's some packages that can only be built on ball
>> because lucatelli/corelli fail every time. azeem has a bunch of those.
>
> Those were number-crunching testsuites and they timed out - I am not
> sure what the relative hardware specs are but it could be that corelli
> is just too slow to run those (likely floating-point heavy) test suites
> in the sbuild time limit.
>
> I haven't looked closer, though.
>
The problem with the octeons is that while they have many cores (32 in
our case), they don't have an FPU. That's why heavy floating-point code
is slow.
I have a swarm at home doing nothing since we moved, if we can find a
place to host it, it seems the best way to go.
--
Aurelien Jarno GPG: 1024D/F1BCDB73
aurelien@aurel32.net http://www.aurel32.net
Reply to: