home *** CD-ROM | disk | FTP | other *** search
- Newsgroups: comp.parallel
- Path: sparky!uunet!europa.asd.contel.com!gatech!hubcap!fpst
- From: Stephen Vavasis <vavasis@cs.cornell.EDU>
- Subject: definition of parallel efficiency
- Message-ID: <1993Jan11.142638.7789@hubcap.clemson.edu>
- Apparently-To: comp-parallel@eddie.mit.edu
- Sender: fpst@hubcap.clemson.edu (Steve Stevenson)
- Organization: Cornell Univ. CS Dept, Ithaca NY 14853
- Date: Sat, 9 Jan 1993 19:03:54 GMT
- Approved: parallel@hubcap.clemson.edu
- Lines: 25
-
-
- I have seen two definitions of "parallel efficiency" in the literature,
- and I am wondering whether one definition or the other has prevailed.
- They are:
-
- (1) par eff = (seq. time of the par. algorithm) / (p * parallel time)
-
- (2) par eff = (seq. time of the best possible sequential algorithm
- for the problem) / (p * parallel time)
-
- If you had a parallel cyclic reduction algorithm, would you rate it
- against a sequential cyclic reduction algorithm or against tridiagonal
- Cholesky factorization?
-
- How about when the underlying problem is more complicated, for example
- solving a boundary value problem? In this case it's not clear how to
- identify the best possible sequential algorithm.
-
- I would like to get this definition straight because I am teaching a
- class on parallelism.
-
- Thanks,
- Steve Vavasis
- vavasis@cs.cornell.edu
-
-