home *** CD-ROM | disk | FTP | other *** search
- Newsgroups: comp.parallel
- Path: sparky!uunet!gatech!hubcap!fpst
- From: kevin-kreeger@orl.mmc.com (Kevin Kreeger)
- Subject: True Bandwidth on CM-2
- Message-ID: <1992Sep15.125814.19175@hubcap.clemson.edu>
- Apparently-To: comp-parallel@uunet.uu.net
- Sender: kevink@orl.mmc.com (Kevin Kreeger)
- Organization: Martin Marietta
- Date: Mon, 14 Sep 1992 22:50:43 GMT
- Approved: parallel@hubcap.clemson.edu
- Lines: 21
-
-
- I've got a question on the true operating Bandwidth of the CM-2
-
- I am attaching to an 8k subset and using a 2-d mesh shape. If I use the
-
- CM_send_to_news(&input,output,0,CM_upward,boolsizeof(char));
-
- exactly how many message startups will be executed. What I mean is that
- If I have a 128x64 shape so I have 8K vitural processors on 8k physical
- processors, and I send a message 8 bits long, do I actually produce
- 8 message startups. Is the true operating bandwidth in this case 1 bit?
-
- Also, since 16 processors sit on a chip, and then there are 12 hypercube
- wire interconnects going off chip, is a mesh embedded in the array, or
- do the messages that go from one chip to the chip physically sitting next
- to it on the board go through hypercube routing (which in some cases
- is longer that 1 unit route, right?).
-
- thanks
- Kevo
-
-