nu.kanga.list.mud-dev

18937: Re: [MUD-Dev] Re: TECH: Distributed Muds

On Thu, 19 Apr 2001 12:49:09 -0400 
Derek Snider <derek@idirect.com> wrote:

> According to J C Lawrence...
>> Derek Snider <derek@idirect.com> wrote:

> That's why I said I wasn't sure that a "Beowulf Cluster" was the
> best solution, but that maybe with the proper configuration and
> software (where by I meant changing the way the system handles IPC
> data) it could be made to work.

<nod>

The base problem is that beowulf's basic IPC mechanism is NFS and
rsh.  Change that and you've fundamentally changed Beowulf.

> You'd probably have to build a system where you take advantage of
> all the machines being in very close proximity (ie: on the same
> rack, or in the same case) and implement some sort of bus-sharing
> system rather than using conventional networking technologies.

Things like PVM have some interesting characteristics.  The research
labs, in particular places like JPL, LLNL, etc are investing heavily
into more MPI/NUMA like clustering architectures.  

> What would really be useful would be a modular super-computer
> design where you could easily "plug-in" more CPUs, RAM,
> hard-drives and network cards.  I know that sounds a little
> far-fetched, but I seem to recall hearing bits and pieces about
> such technologies emerging.

This is moderately standard at the top end.  SGI's brick concept in
particular is aimed squarely at this need (tho for a much higher $$$
tag):

  http://www.sgi.com/origin/3000/

--
J C Lawrence                                       claw@kanga.nu
---------(*)                          http://www.kanga.nu/~claw/
--=| A man is as sane as he is dangerous to his environment |=--
_______________________________________________
MUD-Dev mailing list
MUD-Dev@kanga.nu
https://www.kanga.nu/lists/listinfo/mud-dev