Re: [Condor-users] Pool Of Virtual Boxes

[Ian Cottam <Ian.Cottam@xxxxxxxxxxxxxxxx>]

On 31/03/2011 05:02, "David J. Herzfeld" <david.herzfeld@xxxxxxxxxxxxx>
wrote:

>Hi Ian:
>
>Hopefully everything works as expected. Let met know if you run into any
>issues - I am happy to help/update the source to get things working per
>your expectations (including updating to support the latest version of
>Vbox).
>
>Right now, we run POVB on approximately 400+ machines here are
>Marquette, with various deployment strategies (manual installs,
>DeepFreeze, Altiris, etc.). Our researchers (particularly computational
>chemistry) use the available slots provided by POVB all the time. In
>addition, we have never had any complaints that the systems were running
>slowly due to the underlying virtual machine.
>
>Have a great day,
>DJH
>david.herzfeld@xxxxxxxxxxxxx
>
Thanks David.
[For those who don't know: David is the main author of POVB.]
Our usage case is that, although our main Pool is a dual boot one --
during the day the PCs run Windows for teaching support and at
night/weekends they run Linux/Condor -- our campus has many more PCs that
cannot dual boot for various reasons.

On my first trial with POVB, I set it up pretty much out of the box and
was very impressed that it 'just worked'. Before I knew it two real Condor
jobs had landed on the dual core PC I had used. Now the default setting
are to give Virtualbox half the RAM and all cores. The jobs did not finish
after 24 hours and I knew they were probable only a few hours on our
regular pool machines. I switched to one third memory and one core and did
a better test.

This time: the native run took 30 minutes and the POVB one 37; much better!

I now slightly regret changing two variables, and shall go back and try
with all cores.
What is your advice / experience?

regards
-Ian 
-- 
Ian Cottam
IT Services for Research
Faculty of EPS
The University of Manchester