[M3devel] tinderbox status improvements/diagnosis

[Jay K]

I am using ssh, and just -z1.
5 minutes I don't think so. It takes quite a while..maybe a minute, on a cvs -z3 upd from the root before it transfers any files. But subsequent nearby-in-time runs are faster. Some caching affects but they are surprising -- which cache? How much data is being accessed? I don't know, haven't traced it or anything.
 
 
Yeah some source control is heavyweight but I think many are not.
Git seems to be rapidly gaining in popularity and with that you probably hit "the server" much less.
I worry "distributed source control" like Git would make sharing changes too infrequent, maybe worse than the other side too requent.
 
 
I thought cvsup might be a good solution but haven't tried it yet.
  Only time/interest for so many new things at a time. 
 
 
I'll check with a friend about hosting my weirdo machines.
I see the Mac has good command line control of "scheduled wake".
If all my hardware had "scheduled wake" would be easy.
I only first saw this feature recently on an SGI.
I don't know how common it is.
 
 
Ultimately an Intel Mac with scheduled sleep/wake, and starting up N virtual machines upon startup, either on a schedule or that each wait an hour, two hours, three hours, etc. before doing anything, should be able to substitute for I figure roughly 15 machines -- (Linux + OpenBSD + FreeBSD + NetBSD + Solaris + NT + Darwin) * (I386 + AMD4) + PPC_DARWIN. I had an Intel Mac but returned it and will get a "bigger" (more diskspace) soon.
Heck, just run them all serially in a loop, just one heat spewing machine, should be a good setup.
And then gradually try to add even more esoteric systems (Spin?).
 
 
It kind of feels like cheating but also seems like a good idea.
VMware has even automated the installs of some operating systems, very nice.
Still that leaves PPC, SPARC, MIPS, IA64, Alpha to run on real hardware (though several of those boast virtualization features, I expect it is much harder to get up and running than the x86 virtualization products..)
 
 
 - Jay



----------------------------------------
> Date: Mon, 27 Jul 2009 15:19:56 +0200
> From: wagner at elegosoft.com
> To: jay.krell at cornell.edu
> CC: m3devel at elegosoft.com
> Subject: RE: [M3devel] tinderbox status improvements/diagnosis
>
> Quoting Jay K :
>
>>>>> I've heard cvs is a cpu hog, it's one reason people switch away from
>>>>> it (which I hope we'll do after this release..)
>>>
>>> I'll have to search, initial searches failed.
>>>
>>>> Not that I know... it usually sleeps while waiting for file i/o.
>>>
>>> True.
>>
>> Load average is high on birch, 8.
>> I don't know what this means but I thought it was was supposed to be
>> around 1.
>
> No. The load average is the average number of processes competing for cpu
> within the last one/five/fifteen minutes. See
> http://en.wikipedia.org/wiki/Load_(computing) for details.
>
> In general no process will wait for CPU at all if your load average
> is lower than the number of processors.
>
>> Or is 1 times the number of processors?
>
> o Linux systems count not only runnable processes as the article
> in Wikipedia says.
>
> o Unix systems in general can handle high loads fairly well.
>
> o birch is performing a lot of services: CVS, ssh, ftp, web service,
> trac, mail backup, etc. So a somewhat higher load is expected.
>
>> cvs, by me and hudson, often visible taking a lot, judging by top.
>
> I really think the limiting factor are the disks, unless you are
> using compression and/or encryption (-z, ssh access). But that is
> not inherent to CVS.
>
> You will find that running other SCM systems (like ClearCase,
> Subversion with HTTPS/Apache access, PVCS/Dimensions, ...) will
> put much higher demands to your resources than simple CVS.
>
>> Tinderbox...is causing too much heat in my apartment. After a few
>> green runs I'll have to see about powering the machines on once a
>> week or something for one run and then power back off.
>
> That's understandable.
>
>> Maybe leave just one running daily or in a loop, like LINUXLIB6.
>
>> Later maybe move all x86/amd64 machines to virtual machines and
>> inherit power management of host or scheduled power off/on.
>
> I think we should try to obtain access to a few machines in a
> large data center. Perhaps we can use the HP Partner Virtualization Program.
> I'm not sure if we can get access for an open source project or need
> to make a contract as a company (Elego). We'll investigate that.
>
> Olaf
> --
> Olaf Wagner -- elego Software Solutions GmbH
> Gustav-Meyer-Allee 25 / Gebäude 12, 13355 Berlin, Germany
> phone: +49 30 23 45 86 96 mobile: +49 177 2345 869 fax: +49 30 23 45 86 95
> http://www.elegosoft.com | Geschäftsführer: Olaf Wagner | Sitz: Berlin
> Handelregister: Amtsgericht Charlottenburg HRB 77719 | USt-IdNr: DE163214194
>