[torqueusers] Re: NFS installed Torque/PBS ... security issues ...
jdsmit at sandia.gov
Wed Sep 27 09:38:56 MDT 2006
>> In cluster with NFS context, I assume that folks typically install
>> Torque as
>> root and so that the built binaries reside in a single location across
>> the cluster
>> NFS mounted.
> I don't know if that assumption is accurate. I would assume (and I could
> be wrong, too) that most sites build the TORQUE binaries, and then run
> 'make packages' - dropping and installing the packages locally on each node.
We run from NFS to avoid rpm/build conflicts on nodes that may need to be
rebuilt due to hardware failure / OS problems.
It also helps when upgrading. We have an area where we build multiple
instances of Torque in NFS space ie..
And then point to the correct area via a symlink.
This way if testing goes bad, we do not have to rebuild entirely from
scratch, a previously working version. ( Because we all know even the most
rigorous documentation may miss that 1 change we made to actually get the
stuff to work. :) )
We also keep configuration files in a common NFS area, files like:
This way all nodes share the same files and a change in one file is a change
globally. It makes for much simpler management and debugging when you don't
have to check if every node in a 4500 node cluster has the same config.
This way if a node is powered down/offline for network reasons etc.. We
don't have to worry that it may come up differently from the rest of the
cluster when it is again available.
More information about the torqueusers