home *** CD-ROM | disk | FTP | other *** search
- Newsgroups: comp.parallel
- Path: sparky!uunet!gatech!hubcap!fpst
- From: grendel@fen.arc.nasa.gov (Raymond E. Suorsa)
- Subject: Task allocation for multiprocessors
- Message-ID: <1992Jul23.135001.5315@hubcap.clemson.edu>
- Apparently-To: comp-parallel@ames.arc.nasa.gov
- Sender: usenet@news.arc.nasa.gov
- Reply-To: grendel@windchime.arc.nasa.gov
- Organization: NASA Ames
- Date: Wed, 22 Jul 1992 19:43:40 GMT
- Approved: parallel@hubcap.clemson.edu
- Lines: 31
-
-
- Hello,
-
- I have a little problem some of you should be able to help me with.
-
- I am working with a computer vision problem domain which may be
- divided into M independent tasks with each task requiring a variable
- amount of time T_i, where the total problem is solved in T seconds of
- processor time:
-
- T=sum_{i=1}^M T_i
-
- The optimal allocation of these tasks onto a N processor machine,
- where M > N, is an NP-complete problem.
-
- There are several ways to do a good allocation by first sorting the
- set of M tasks, and assigning the longest running tasks first. This
- helps reduce the balancing errors as the end of the sorted list is
- approached.
-
- I have found many papers concerning task allocation using graph
- methods. I am looking for a simpler treatment closer to this problem
- type. (i.e. What is the Best Choice for M given N and the minimum and
- maximum expected T_i.)
-
- Thanks,
-
- __o Ray Suorsa,
- \<, grendel@windchime.arc.nasa.gov,
- ()/ () NASA/Ames (415)604-6334
-
-