[Adium-devl] Hardware plans
timber at lava.net
Thu Aug 23 15:21:17 UTC 2007
On Aug 23, 2007, at 3:12 AM, Evan Schoenberg wrote:
> On Aug 22, 2007, at 11:57 PM, Peter Hosey wrote:
>> The questions, then, are:
>> 1. Could distcc on a cluster of minis make up the loss of clock
>> cycles with pure parallelism, being able to compile 12 source files
>> at a time rather than four?
I don't really think dist-cc is all that useful, although it's worth
investigating. I think a new mac mini would be Fast Enough.
>> (Keeping in mind, of course, that it will take some fraction of a
>> second longer for each compiler process to finish, and that that
>> fraction must currently be multiplied by 608—this being the sum of
>> 500, which is the number of .m files in the Adium project, and 108,
>> which is the number of .m files in the AIUtilities project.)
>> 2. Would 2 GiB of RAM be enough for each distributed-build node?
> I don't know the answers (though my experiences with distcc have
> never left me particularly impressed) but add two more questions:
> 3. For purposes other than building, is having multiple computers
> more or less useful than having a single, powerful one?
> 4. Is managing multiple systems worth the added management burden
> versus a single system?
Yes. You get redundancy, which is very important.
I talked with the build/release folks here at Mozilla, and they said
they've given up buying Xserves (we have a couple). For build
machines, it's easier to just have a bunch of mac mins. Buildbot
allows you to have multiple slave machines, so when a command comes in
it can pick one out of the pool.
We can even write steps for it to automate our release process, so
someone with the right ssh key or password or whatever could kick off
a release build and not have to do all that work by hand.
You know, I'm not sure why we're planning on running things on these
machines other than Mac OS X only things (e.g. Jabber server). It
seems like a waste of those resources.
My recommendation is to figure out what we can purchase TODAY. One or
two mac minis would be fine. Waiting for something newer over the
horizon means we aren't getting anything done. Once we see our
resource utilization, and what we're actually doing with the machines,
then we can buy more as they are warranted. Just buying one super
powerful Xserve to play around with sounds like a horrendously bad
idea to me, the more i think about it.
-------------- next part --------------
An HTML attachment was scrubbed...
More information about the Adium-devl