home *** CD-ROM | disk | FTP | other *** search
- Xref: sparky comp.parallel:3038 comp.theory:2918
- Newsgroups: comp.parallel,comp.theory
- Path: sparky!uunet!gatech!hubcap!fpst
- From: vram@cc.gatech.edu (Venkataraman Ramanathan)
- Subject: Re: Vertex disjoint routing on Hypercube
- Message-ID: <1993Jan23.153720.15389@hubcap.clemson.edu>
- Apparently-To: comp-parallel@gatech.edu
- Sender: news@cc.gatech.edu
- Reply-To: vram@cc.gatech.edu (Venkataraman Ramanathan)
- Organization: College of Computing
- Date: Sat, 23 Jan 1993 15:07:29 GMT
- Approved: parallel@hubcap.clemson.edu
- Lines: 26
-
- Let's say V1 and V2 differ in n bits. By an easy transformation, we could
- get V1 to origin and V2 to V. Now, the problem is to find n disjoint routes
- from O (origin) to V. For a 8 dim difference,
-
- 1 2 3 4 5 6 7 8
- 2 3 4 5 6 7 8 1 (1 rotated)
- 3 4 5 6 7 8 1 2 (2 rotated)
- ...
- 8 1 2 3 4 5 6 7 (7 rotated, n-1 in general).
-
- consider the scheme above. There can be atmost n routes.
- For each route, flip the bits (move in that dimension) according to
- the sequence defined by the row.
-
- First route is : take dim 1, 2 ... (1 row)
- Second route is: take dim 2 first, 3 next, and so on..
- And so on..
-
- Proof is easy. This rotate-router is very effective. Fill your route number in,
- increment and do mod + 1, keep flipping these bits..: that's all.
- -
- Venkat
- (vram@cc)
- PS. I got 10/10 on this homework problem with Dr.Jim Burns.
- PPS. Ah! this didnt show up when i posted this through 're'. now it's ok!
-
-