home *** CD-ROM | disk | FTP | other *** search
- Path: sparky!uunet!cis.ohio-state.edu!zaphod.mps.ohio-state.edu!caen!nic.umass.edu!dime!dime.cs.umass.edu!moss
- From: moss@cs.umass.edu (Eliot Moss)
- Newsgroups: comp.object
- Subject: Re: Is GC needed?
- Message-ID: <MOSS.92Jul25091639@ibis.cs.umass.edu>
- Date: 25 Jul 92 13:16:39 GMT
- References: <DJOHNSON.92Jul6152406@mycroft.ti.com> <1992Jul8.012152.28375@clam.ssw.de>
- <1992Jul10.165626.7864@visix.com> <4567@balrog.ctron.com>
- <AXEL.92Jul24154632@avalanche.cs.tu-berlin.de>
- Sender: news@dime.cs.umass.edu
- Reply-To: moss@cs.umass.edu
- Organization: Dept of Comp and Info Sci, Univ of Mass (Amherst)
- Lines: 31
- In-reply-to: axel@cs.tu-berlin.de's message of 24 Jul 92 14:46:32 GMT
-
- >>>>> On 24 Jul 92 14:46:32 GMT, axel@cs.tu-berlin.de (Axel Mahler) said:
- Axel> Nntp-Posting-Host: avalanche.cs.tu-berlin.de
-
- Axel> Have there been any attempts to employ multi-processor shared-memory
- Axel> computer architectures for solving the GC problem ? Why not have a
- Axel> dedicated GC processor that frees the application from worrying about
- Axel> memory management ? I've not given the idea a great deal of thought, so
- Axel> it might actually be stupid or unfeasible. Comments ?
-
- I regret I have no references handy, but yes indeed it is an idea that has
- been worked on some. First, collection concurrent with computation is a
- venerable problem, studied by Dijkstra, among others, with a number of
- proposals, some of them really implemented. Second, firmware or other
- distinctive hardware support for concurrent or overlapped gc (or parts of gc)
- has also been explored a bit. Doing hardware always has the problem of getting
- people to accept it, and the steep curve to get it implemented, integrated,
- etc. Third, gc within a concurrent shared memory system, either by separate
- dedicated processes/threads (could be dedicated cpus) or by threads that
- alternate between computation and collection are things that have been and are
- being investigated. So, no, it's not a stupid idea. The hard part is getting
- the overhead of coordination low enough to make it pay, unless you're doing
- systems where predictability is more important than absolute resource
- consumption.
- --
-
- J. Eliot B. Moss, Associate Professor
- Department of Computer Science
- Lederle Graduate Research Center
- University of Massachusetts
- Amherst, MA 01003
- (413) 545-4206, 545-1249 (fax); Moss@cs.umass.edu
-