Re: [linux-linus test] 104237: regressions - FAIL

[Julien Grall <julien.grall@arm.com>]

Hi,

On 19/01/2017 20:15, Boris Ostrovsky wrote:
>
>> So this appears to be a pretty slow, 2-core box:
>>
>> [ 0.049778] SMP: Total of 2 processors activated (96.00 BogoMIPS).
>
> For comparison, my (old-ish) phone is 4 x 26 BogoMIPS. So it's
> theoretically faster than this server.

The ARM32 platforms we have in the colo are not server but development 
board :). Some of them are based on early revision of cores and can be 
really really really really ... slow.

>
> -boris
>
>>
>> Julien --- how does 2.5 hours for kernel build sound to you? (make -j4
>> all modules)

I would not be surprised that it takes 2.5 hours for a make allmodules 
on both the Cubietruck and Arndale.

I have looked at the Linux-3.18 test flights, I know that this thread is 
linux-linus but not other flight log can be found. So surprisingly the 
time betwen the last two flights of Linus 3.18 has doubled (see [1] and 
[2]).

The major difference is a Cubietruck is used on the latest version and a 
Arndale on the previous one. I am be surprised that the Cubietruck is 
slower than the Arndale. Oh well..

The interesting bit is the build time on the Cubietruck is very close to 
the threshold. It is likely that the latest kernel have more modules to 
build which increased the time to build.

So I can see 2 solutions:
	1) Increase the timeout
	2) Only build the kernel on the Arndales. Though they are known to be 
unreliable in the colo :/

Any opinions?

Cheers,

[1] 
http://logs.test-lab.xenproject.org/osstest/logs/104271/build-armhf-pvops/info.html
[2] 
http://logs.test-lab.xenproject.org/osstest/logs/103983/build-armhf-pvops/info.html

-- 
Julien Grall

_______________________________________________
Xen-devel mailing list
Xen-devel@lists.xen.org
https://lists.xen.org/xen-devel