home *** CD-ROM | disk | FTP | other *** search
- Newsgroups: comp.parallel
- Path: sparky!uunet!gatech!hubcap!fpst
- From: mln@sunkj.larc.nasa.gov (Michael L Nelson)
- Subject: PVM and Intel iPSC/860 (was Re: PVM vs. Express)
- Message-ID: <1992Nov19.132428.9852@hubcap.clemson.edu>
- Keywords: PVM, Hypercube, iPSC/860
- Sender: fpst@hubcap.clemson.edu (Steve Stevenson)
- Nntp-Posting-Host: sunkj.larc.nasa.gov
- Organization: NASA Langley Research Center, Hampton, VA USA
- References: <1992Nov13.125101.20219@hubcap.clemson.edu> <1992Nov16.131740.5002@hubcap.clemson.edu> <1992Nov16.190340.7465@hubcap.clemson.edu> <1992Nov18.130411.25733@hubcap.clemson.edu>
- Date: 18 Nov 1992 23:15:29 GMT
- Approved: parallel@hubcap.clemson.edu
- Lines: 48
-
- Tom Crockett (ICASE) referenced my work with PVM and the Intel iPSC/860
- and I received a couple of requests about it so I thought I would
- post a summary.
-
- It currently runs as PVM 2.4.1. I have been waiting for PVM 3.0 to come
- out to perform a "cleaner" modification.
-
- I have modified the PVM daemon and its associated library so that an
- Intel Hypercube may be more successfully integrated into a distributed
- environment.
-
- Cube programs are able to link the PVM library and have PVM calls on the
- cube. The daemon that controls these program does not reside on the cube
- or on the SRM. This daemon is moved to another machine. In practice,
- this has been a workstation. So far, when a workstation is performing as
- a PVM front end for the cube, that WS can not support both the modified and
- unmodified version of the daemon concurrently.
-
- The daemon that runs on this workstation has been modified so that it
- will be controlling programs on an iPSC/860 somewhere. The cube programs
- communicate with the daemon through the cube's service node (which has
- its own IP address) and not through the SRM.
-
- It is important to note that this method does involve the creation of
- a "stub" program on the workstation side to correctly fire off the
- appropriate cube programs, but these "stub" programs are only to correctly
- load the program onto the cube are not used past this point (ie, they perform
- no "conversion" of messages).
-
- It is also important to note that this only provides the capability to have
- PVM messages on the cube and removes the bottleneck of running the PVM
- daemon and the previously necessary conversion programs on the SRM. It is up
- to the programmer to write sufficiently clever code to take advantage of this
- added capability.
-
- Please note that PVM on the cube is *not* meant to replace the native
- message passing library. Again, the aim is to allow the Hypercube to be a
- node in a distributed environment.
-
- I will make the 2.4.1 version available anonymous ftp if there is sufficient
- interest. I will have a 3.0 version as soon as possible.
-
- Michael Nelson
- NASA-Langley Research Center
- M.L.Nelson@LaRC.NASA.GOV
-
-
-
-