home *** CD-ROM | disk | FTP | other *** search
- Path: sparky!uunet!think.com!ames!agate!darkstar.UCSC.EDU!osr
- From: rlarowe@chpcchpc.org (Rick LaRowe)
- Newsgroups: comp.os.research
- Subject: Re: Difference between DYNAMIC and ADAPTIVE load balancing???
- Date: 6 Nov 1992 21:03:54 GMT
- Organization: Center For High Perf. Computing of WPI; Marlboro Ma
- Lines: 45
- Approved: comp-os-research@ftp.cse.ucsc.edu
- Message-ID: <1demjqINNl34@darkstar.UCSC.EDU>
- References: <1d9guvINN6rp@darkstar.UCSC.EDU> <1dbi73INNm41@darkstar.UCSC.EDU> <1dcbqrINNol@darkstar.UCSC.EDU>
- NNTP-Posting-Host: ftp.cse.ucsc.edu
- Originator: osr@ftp
-
-
- In article <1dbi73INNm41@darkstar.UCSC.EDU> xander@paramount.nikhefk.nikhef.nl (Xander Evers) writes:
-
- >The terms dynamic scheduling and adaptive scheduling are sometimes used
- >in the literature in an inconsistent manner. A dynamic schedular takes
- >into account the current state of affairs as it perceives it in the system.
-
- It appears to me that the difference between a dynamic and an adaptive
- algorithm or policy is largely a question of whether you use the latest
- buzzword or not. Say I've got two "dynamic" algorithms A and B, each
- with a couple of parameters that affect their behavior. Based upon
- some of the responses to the original post on this topic, lots of folks
- would not consider these "adaptive" policies, even if they "react" to
- the system state. But if I go and create policy C that uses some new
- parameter values to select between policies A and B, it suddenly
- becomes "adaptive", because the "basic control flow of the policy
- changes in reaction to system state." But had I originally gone out
- and implemented the basic functionality of policy C without mentioning
- policies A and B, my description to it would have sounded alot like the
- description I just gave for A and B (namely, they have parameters that
- affect the way the policy behaves given certain system conditions).
-
- So my question to the folks who claim there is really a difference;
- where do you draw the line? When is a reaction to a change in system
- state part of the basic algorithm, and when is it a change in the basic
- control flow of the policy? In particular, I'd be curious to know
- whether the NUMA memory management policy we (Carla Ellis and I)
- studied in our SOSP '91 paper is dynamic or adaptive. We called it
- dynamic (adaptive wasn't such a hot buzzword when we started that work
- 8-)), but given the extent to which its behavior can change (sometimes
- replicating, sometime migrating, sometimes using remote reference
- depending on the workload encountered), I can imagine calling in
- adaptive.
-
- Anyway, if there is a difference between dynamic and adaptive, I'd
- certainly like to see a clearer definition.
-
- Rick LaRowe
- ------------------------------------------------------------------------------
- Center for High Performance Computing internet: rlarowe@chpc.org
- Worcester Polytechnic Institute rlarowe@wpi.edu
- Suite 170 phone: (508) 624-7400 x610
- 293 Boston Post Road West (508) 831-5837 (campus)
- Marlborough, MA 01752 (508) 624-6354 (fax)
- ------------------------------------------------------------------------------
-