Newsgroups: comp.parallel.pvm
From: anton@aix.kingston.ibm.com (Anton Prenneis)
Reply-To: anton@mhv
Subject: Re: Feasibility of PVM Under (IBM) SMP Hardware?
Organization: IBM POWERparallel Workload Management
Date: 16 Apr 1996 20:05:25 GMT
Message-ID: <1996Apr16.160413@aix.kingston.ibm.com>

(Posting for Honbo Zhou)

> Newsgroups: comp.parallel.pvm
> Subject: Re: Feasibility of PVM Under (IBM) SMP Hardware?
> References: <317287D2.41C6@ntsg.umt.edu>
> Organization: Cornell Theory Center, Cornell University
> Keywords:
>
> In article <317287D2.41C6@ntsg.umt.edu> Joe Glassy <glassy@ntsg.umt.edu> writ
es:
>I am relatively inexperienced in PVM culture and have questions...
>We are considering purchasing an IBM J-30 SMP. Could anyone offer
>opinions on the general applicability or feasibility of running PVM
>apps on this type of K way SMP? I'm particularly interested in how
>one maps particular CPU nodes to given execution instances,
>assuming one can do this via startup directives,or that the PVM
>daemon "automatically" does this.
>

PVM has introduced a machine (ARCH) type "RS6KMP" for IBM SMP machines.
When you run PVM on IBM SMP, you don't need to do anything, PVM will
know "automatically" that you are running on a SMP machine and it will
also automatically assign (with help of the multithreaded AIX 4.1 OS)
PVM tasks to different processors on the SMP in a loadl-balanced fashion.

Also, the message passing bandwidth of PVM (I measured) on IBM SMP is 2+ times
faster than that of on SUN SMP because IBM SMP's cache augumented BUS
structure provides faster virtual shared memeory-to-memory copy.

Good Luck,

Honbo Zhou

>Has anyone had any experience in running the IBM SMP in "single image"
>mode, and if so--does this imply that the user has shell level control
>over how simple command shells are mapped to particular CPU nodes.
>
>If PVM is considered a un-feasible or sub-optimal scheme for deploying
>"parallel" jobs on an SMP, does MPI generally offer conceptual and/or
>structural advantages?
>
>thanks in advance for any opinions...
>
>regards,
>joe glassy
>
>--
>[   "Everyone has to believe in something. I believe I'll go fishing."]
>[ -- (H.D. Thoreau) --------------------------------------------------]
>Joe Glassy, Dir. Software Devel, NTSG     Internet: glassy@ntsg.umt.edu
>School of Forestry, Univ. of Montana        Office Voice (406) 243-2057
>Missoula, MT  59812                         Office FAX   (406) 243-4510



-A
-- 
 ____________________________________________________________________________
 Anton A. Prenneis, Jr. | "Computers in the future may weigh no more than 1.5
 IBM RS/6000 Division   |  tons."  - Popular Mechanics, 1949.
 POWERparallel Systems  | - - - - - - - - - - - - - - - - - - - - - - - - - -
 Poughkeepsie, N.Y.     | "I think there's a world market for about five
                        |  computers."  - Thomas J. Watson, 1943.


