[Beowulf] need for an advice on nfs and diskless clients
mkinet at ulb.ac.be
Thu Feb 21 00:57:50 PST 2008
> Would it be feasible to use "ln" to create symbolic links to
> populate /tftproot/192.168.1.N, for each node N, from say /tftproot/
> generic ?
I tried that trick but didn't work. Apparently, symbolic links are not
enough to link files at boot time.
What I was wondering is, if the directories /tftpboot/192.168.1.N
contains sub-directories that are indentical for every node, and that
are not supposed to be modified by the nodes, why not put them in a
directory /tftpboot/generic and make the node mount the it from there?
The problem is I have no idea of which directories have to be local,
and which could be shared...
Thanks for the answer anyway.
Université Libre de Bruxelles
Physique Statistique et Plasmas, CP 231
Campus Plaine - Boulevard du Triomphe,
> On Wed, Feb 20, 2008 at 5:26 AM, Maxime Kinet <mkinet at ulb.ac.be>
> I'm setting up a cluster, made of diskless workstations, booting
> through NFS. For that purpose I created a copy of the node's
> filesystem in a /tftpboot directory. That is, I have a full
> filesytem for each single node : /tftpboot/192.168.1.1, /tftpboot/
> 192.168.1.2, /tftpboot/192.168.1.3, etc.
> I know that there are several different options to make several
> clients mounting their filesystems, and that this one might not be
> the most efficient one, but I choosed that one just because I knew
> already how to do it. Now everything works fine, except that I would
> like to spare some diskspace on the master partition. Since each
> filesystem is around 500 Mb, I quickly reach more than 10Gb just for
> the /tftpboot directory.
> A quick solution I founded is to mount the largests directory (which
> are the same for each node) of the / (namely /lib and /var) from a
> separate common directory. Could this cause some problem in the way
> the node's are working? Is there any other easy/fast way to reduce
> the amount of data contained in the /tftpboot/192.168.1.* directories?
> I also heard about ClusterNFS which would do exactly what I need,
> but then I would have to start everything from the beggining and I
> want everything to work as soon as possible.
> Thanks a lot for answers/advices,
> Maxime Kinet
> Université Libre de Bruxelles
> Physique Statistique et Plasmas, CP 231
> Campus Plaine - Boulevard du Triomphe,
> 1050 Bruxelles.
> Beowulf mailing list, Beowulf at beowulf.org
> To change your subscription (digest mode or unsubscribe) visit http://www.beowulf.org/mailman/listinfo/beowulf
-------------- next part --------------
An HTML attachment was scrubbed...
More information about the Beowulf