[Beowulf] Diskess cluster dictribution

[rf at q-leap.de]

>>>>> "Patrick" == Patrick Kirby <otter at rockisland.com> writes:

Hi Patrick,

    Patrick> Hello, We have been using Pelican HPC for building a 14
    Patrick> node cluster.  The slaves are diskless and have AMD FX 8120
    Patrick> CPU's with 8GB ram.  We are running the Evolutionary
    Patrick> Objects genetic program on the cluster.

    Patrick> We have been having trouble with the cluster and would like
    Patrick> to switch to a different cluster software.  The problems
    Patrick> seem to stem from the fact that Pelican is not tolerant of
    Patrick> any interruption is communication with slaves.  I would
    Patrick> appreciate suggestions for a fault tolerant replacement for
    Patrick> Pelican HPC that would be appropriate for this project.
    Patrick> Thank you,

without knowing of what nature exactly your communication problems are,
it's hard to suggest something that would solve these problems.

But since you asked about a suggestion for a replacement of Pelican HPC
(which is end-of-life anyhow) you might find Qlustar (qlustar.com) an
attractive replacement. It also supports Debian as a compute platform
as well as diskless cluster nodes out of the box, just like
Pelican. Since the node OS is running in RAM (rather than with NFS root)
it is definitely much less susceptible to network communication
problems. It's easy to try it out: A head-node + virtual demo cluster is
installed in less than an hour. Qlustar is free for pure academic and
non-commercial use and very attractively priced for other cases.

Best,

Roland

----
Roland Fehrenbacher, PhD
Founder/CEO
Q-Leap Networks GmbH
Tel. : +49(0)7034/277620
EMail: rf at q-leap.com
http://www.q-leap.com / http://qlustar.com