home *** CD-ROM | disk | FTP | other *** search
- Newsgroups: comp.parallel
- Path: sparky!uunet!destroyer!gatech!hubcap!fpst
- From: nikhil@crl.dec.com (R.S. Nikhil)
- Subject: Re: EM-4(?) Question
- Message-ID: <1992Sep4.154551.26022@crl.dec.com>
- Keywords: interconnect
- Sender: news@crl.dec.com (USENET News System)
- Organization: DEC Cambridge Research Lab
- References: <1992Sep4.015211.25673@cis.ohio-state.edu>
- Date: Fri, 4 Sep 1992 15:45:51 GMT
- Approved: parallel@hubcap.clemson.edu
- Lines: 24
-
- In article <1992Sep4.015211.25673@cis.ohio-state.edu>, martens@python.cis.ohio-state.edu (Jeff Martens) writes:
- > In a recent report posted to the net, David Kahaner said that the
- > interconnect used in one (or all?) of the EM-3/4/5 family is the
- > "processor omega with deadlock-free mechanism."
- >
- > What exactly is the processor omega? Is it similar to the
- > homogenous circular shuffle (aka single-stage cube, used in the
- > Man-Yo)?
-
- I think it means the following:
-
- In a conventional picture of an omega network, you see an interconnection
- of switching nodes with N inputs on the left and N outputs on the right;
- a processor is typically connected to an input on the left column and the
- corresponding output on the right column.
-
- In the EM-4, the inputs on the left are just wired around to corresponding
- outputs on the right. A processor hangs off each switching node. Each
- switching node is a 3x3 crossbar; 2 inputs and outputs are used for the
- omega interconnection and the 3rd input and output is connected to the
- processor. In fact, the 3x3 switch is on the processor chip itself.
-
- Nikhil (nikhil@crl.dec.com)
-
-