[Beowulf] ethernet switch, dhcp question

[Robert G. Brown]

On Mon, 4 Oct 2004, JOHNSON,PAUL C wrote:

> All:
> 
> Im fairly new to beowulf clusters so please excuse the question if 
> it is trivial.  Ive installed mpich on several computers and have 
> run several programs but the performance seems a little slow.  All 
> the computers in my lab are connected directly to the campus 
> network.  Would I see an increase in performance if I instead had 
> slaves connected through a switch in my room connected to a master 
> computer using dhcp to assign ip's?

Quite probably, depending on how your campus is networked.  A local
switch is the preferred method of building a cluster.  If you are using
a 100 BT network and only have a small cluster, a 100BT switch is so
cheap it is almost a non-issue.  Even if you DO leave them connected to
the campus network, if you do this by interconnecting your switch and
the campus network you'll likely see a performance increase.

Putting them on a private network gives you an even quieter networking
environment and better control over the network, but yes, it will make
you learn a whole bunch of things (e.g. DHCP/PXE and more) to get it
right.  If you go this route, I'd strongly urge that you get PXE-capable
network cards and set up fully automated installation and booting at the
same time.  You'll spend a month learning all sorts of complicated
networking, but at the end of it you'll REALLY save time on
installation, operation, and so forth and your cluster will be upwardly
scalable in size with very little additional effort.

   rgb

> Thanks for any help,
> Paul
> 
> --
> JOHNSON,PAUL C
> 
> _______________________________________________
> Beowulf mailing list, Beowulf at beowulf.org
> To change your subscription (digest mode or unsubscribe) visit http://www.beowulf.org/mailman/listinfo/beowulf
> 

-- 
Robert G. Brown	                       http://www.phy.duke.edu/~rgb/
Duke University Dept. of Physics, Box 90305
Durham, N.C. 27708-0305
Phone: 1-919-660-2567  Fax: 919-660-2525     email:rgb at phy.duke.edu