home *** CD-ROM | disk | FTP | other *** search
- Path: sparky!uunet!opl.com!hri.com!spool.mu.edu!sdd.hp.com!saimiri.primate.wisc.edu!caen!umeecs!adirondacks.eecs.umich.edu!jrexford
- From: jrexford@adirondacks.eecs.umich.edu (Jennifer Lynn Rexford)
- Newsgroups: comp.arch
- Subject: Non-uniform traffic in message-passing systems
- Message-ID: <1993Jan27.204625.17541@zip.eecs.umich.edu>
- Date: 27 Jan 93 20:46:25 GMT
- Sender: news@zip.eecs.umich.edu (Mr. News)
- Organization: University of Michigan EECS Dept., Ann Arbor, MI
- Lines: 23
-
-
- Does anyone know of any papers that characterize realistic
- communication patterns in message-passing systems? In particular,
- how often are various communication constructs (such as many-to-one,
- one-to-many, and others) used over a wide range of applications?
-
- I am looking at non-uniform traffic patterns on point-to-point,
- message-passing networks (such as k-ary n-cubes). The shared-memory
- community has models of non-uniform traffic in terms of a ``hot-spot''
- memory, that can cause congestion trees in multistage interconnection
- networks. It seems that a ``hot'' node (or set of nodes) in a message
- passing environment (or a distributed shared memory environment) would
- be analogous.
-
- Any thoughts/references?
-
- I do have one reference ("Hot-Spot Contention in Binary Hypercube
- Networks," by Dandamudi and Eager, IEEE TOC, Feb. 1992, pp. 239-244).
-
- Thanks.
-
- -- Jennifer
- jrexford@eecs.umich.edu
-