replacement builder needed

Jarod Wilson jarod at wilsonet.com
Tue Oct 25 15:49:54 CEST 2011


On Oct 25, 2011, at 9:27 AM, Richard Shaw wrote:

> On Tue, Oct 25, 2011 at 8:15 AM, Jarod Wilson <jarod at wilsonet.com> wrote:
>> On Oct 24, 2011, at 3:07 PM, Dominik 'Rathann' Mierzejewski wrote:
>> 
>>> On Sunday, 23 October 2011 at 20:02, Richard Shaw wrote:
>>>> Since no one has stepped up yet I have a proposal but I don't know how
>>>> easy it will be to accomplish.
>>>> 
>>>> I'm willing to offer up my desktop machine with the following requirements:
>>>> 
>>>> 1. At least 5 to 10 others in total do the same
>>>> 2. Building packages would be assigned to all machines in a round
>>>> robin approach.
>>>> 3. I can shut down the builds by window or manually. I do occasionally
>>>> use my machine for gaming.
>>>> 
>>>> Another option in addition to the above would be to allow uploading of
>>>> packages from contributors machines to buildsys. I would suggest
>>>> making it possible from cvs/make where the packages would get built on
>>>> the local machine and then be uploaded.
>>> 
>>> I can set up a builder on my home server, if necessary. However my uplink
>>> is only 512kbps at the moment and the server is a 64bit dual-core Atom,
>>> so not exactly a speed demon. :)
>> 
>> I tried running a similar atom box as a builder at one point. It was horrendous,
>> the thing kept choking on builds, took ages to complete anything, often got
>> itself completely hung due to excessive swapping, would oom-kill, etc. But if
>> you had tons of RAM in it (read: more than 4GB), it *might* be okay. I'd be
>> more inclined to roll with what Adrian is cooking up. :)
> 
> As a point of reference, I get decent build times on my AMD X2 Kuma
> 2.7GHz machine with 4GB of memory.

The current host is actually right about the same specs. The atom might have
only had 2GB, not certain. But the cpus were definitely overwhelmed.

> Mythtv (after initial package caching in mock) took about 25 minutes
> Building the kernel takes right at 2 hours.

Note that plague will, iirc, queue both the x86_64 and i686 builds at the same
time. And so plague thought the dual-core atom with hyperthreading turned on
should be doing two simultaneous -j4 builds. It was less bad with two -j2 builds,
but still pretty bad. The atom is flat-out a terrible cpu for anything even remotely
computational. A dual-core Athlon X2 handles the load just fine.

> I think a quad core with close to 3GHz speeds would be ideal :)

Remember, clock speed ain't everything. I'd rather have a low-power 2GHz
quad-core i7, which would still trounce an Athlon X2.

-- 
Jarod Wilson
jarod at wilsonet.com



More information about the rpmfusion-developers mailing list