On Thu, 27 Jan 2005, Torsten Harenberg wrote:
> Dear all,
>
> I would like to bring up a general question and hope you can give us some
> advice:
>
> We're operating a cluster running 1024 Opterons. As OS SuSE 9.0 prof.
> 64bit is installed (but nearly all 32bit libs are also there). PBS is used
> as a batch system.
>
> I want to have include that machine (at least in part) into LCG.
>
> For several reasons, I cannot change the OS from SuSE to SL. But as we for
> example run ATLAS-Software on the machine w/o problems, that shouldn't
> make too much trouble.
>
> Is there any proposed way how to proceed? I would assume that all the WN
> specific RPMs would have to be installed (which would be no problem), but
> does everything needed is situated under /opt/lcg, /opt/edg, or /etc
> afterwards, so that it wouldn't affect the SuSE system? Has anybody tried
> something like this and how much efford was this?
>
> I remember that Dresden reported to have Debian systems configured to act
> as LCG nodes, but if I remember correctly, this was a very big efford.
>
> Any advice, hint, whatever is useful!
My experience in this, having just installed a Debian cluster with LCG2,
is that the tar-wn enviroment seems fairly portable. The more complex node
types probably need to run on SL3 (or similar), but these are just a few
machines.
My suggestion:
1. Install the tar-wn on the cluster nodes.
2. Install lcg2 ce/se/whatever on sl3 computers that can talk to the batch
system (qsub/qstat etc or equivalents).
3. Hope that the software the users are running do not have weird
redhat-specific dependancies.
/Mattias Wadenstein
|