[Beowulf] A fairly simple question

Robert G. Brown rgb at phy.duke.edu
Sat Aug 13 09:04:53 EDT 2005


Mark Hahn writes:

>> Simple solution is single pc with virtual PCs. it works properly as 
> 
> I disagree.  you can always run multiple MPI worker tasks on a single 
> CPU - you won't see speedup, of course, but you can work on your programming
> logic pretty reasonably.  I don't see why virtual PCs would offer any 
> advantage over this.  the next step is to use a few of real PCs as RGB
> suggests - not only do you start to see speedup, but you can play with 
> the reasons why speedup is not perfect.  however, I think just two PCs
> is a useful cluster - there's no big reason to have a separate master,
> and even with 2 boxes, you can learn.


...although to see a nonlinear saturation of speedup -- a scaling
"curve" -- you need at least three point, which is why I suggested four
nodes (there is a long and possibly silly tradition of building clusters
with power-of-two numbers of nodes).

It IS possible to do this even with one node, but only with a toy
application -- one that substitutes null delays e.g. sleep() for real
work between real or simulated communication steps.  sleep() executes in
parallel even on a single processor, obviously.  However, you can't see
"real" parallel speedup on a single processor, or play much with
communications to computation ratios with only two.

    rgb

> 
> _______________________________________________
> 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 --------------
A non-text attachment was scrubbed...
Name: not available
Type: application/pgp-signature
Size: 189 bytes
Desc: not available
URL: <http://www.clustermonkey.net/pipermail/beowulf/attachments/20050813/042e7e08/attachment-0001.sig>
-------------- next part --------------
_______________________________________________
Beowulf mailing list, Beowulf at beowulf.org
To change your subscription (digest mode or unsubscribe) visit http://www.beowulf.org/mailman/listinfo/beowulf


More information about the Beowulf mailing list