[PLUG] surprising performance on my HNOW nodes

[Russell Senior]

>>>>> "Derek" == Derek Loree <drl at drloree.com> writes:

Derek> Does it have to be a workstation (I have a couple of servers
Derek> with spare cycles as well as workstations)?

Either one.

Russell>  ... and all I want is a few trillion cycles. ;-)

Derek> What type of processor(s) are you looking to steal cycles from?

It shouldn't matter to me.  I've currently got everything from P2
300MHz to P4 2.53GHz working, but recompiling the slave is easy.  As
reported elsewhere in this thread, P3s seem to do better than P4s on
the code.  It might need tweaking for a 64-bit architecture.

Russell> Basically, you'd run a slave process on your box (niced to
Russell> taste) which connects to my master program.  The latter hands
Russell> out a job and the former computes random walks.  I haven't
Russell> worked out how to manage it yet.  I either need a ssh login
Russell> on the box, or I could provide you the slave program and we
Russell> could set up some kind of watchdog to restart it if the
Russell> master dies.  I am still debugging a few bits and the master
Russell> might have to restart occasionally, necessitating
Russell> reconnection.

Derek> Either scenario could be arranged.

The ssh method is preferable from my point of view at the moment,
since the other arrangement is a little less flexible (and doesn't
exist at the moment), but I can understand how people might have
sensitivities about that.

Russell> It might be good for an acknowledgement in a journal
Russell> article. ;-)

Derek> What license are you going to release the HNOW under?

Well, the HNOW is just a collection of boxes.  The master/slave
software I've written, though of limited application, can be GPL'd.

Derek> I think I can spare a few trillion niced cycles.

Cool!  Thanks.

I've got more testing and debugging to do.  I'll probably be ready to
get started next week sometime.


-- 
Russell Senior         ``I have nine fingers; you have ten.''
seniorr at aracnet.com