home *** CD-ROM | disk | FTP | other *** search
- Newsgroups: comp.sys.isis
- Path: sparky!uunet!srvr1.engin.umich.edu!batcomputer!cornell!ken
- From: ken@cs.cornell.edu (Ken Birman)
- Subject: Re: Challenges?
- Message-ID: <1993Jan22.154402.24721@cs.cornell.edu>
- Organization: Cornell Univ. CS Dept, Ithaca NY 14853
- References: <1993Jan19.140117.1199@cs.cornell.edu> <8140004@otter.hpl.hp.com>
- Date: Fri, 22 Jan 1993 15:44:02 GMT
- Lines: 37
-
- In article <8140004@otter.hpl.hp.com> pdh@otter.hpl.hp.com (Paul Harry) writes:
- >(discussion of n-n communication issues)
-
- Paul, this is an interesting issue.
-
- First, having looked a bit at the problem, I am convinced that we could
- support a fairly efficient n-n communication tool, using abcast to
- send messages into each group, introducing a coallator for the
- duplicate messages coming in and the duplicate replies back to the
- origin group, and using a multicast to send the reply to the origin
- group instead of the current scheme, which sends to only the origin
- process. None of this would be hard.
-
- However, two observations. First, the cost of such a scheme will be
- high. Second, there may be a cheaper alternative with the same reliability
- -- for example, the scheme that Robert Cooper, Barry Gleeson and I
- wrote up in the TR on "process group semantics" (unpublished). So, I
- can see merit in supporting n-version programming and hence n-n
- communications, but I am not 100% convinced that the need extends to
- other fault-tolerance situations.
-
- Question: can people point to applications in which response time is
- so critical that detecting the failure and then restarting the
- computation is too slow, and hence you need to use an n-n scheme
- instead of a primary-backup scheme? I know of a few such things,
- but my theory is that you can localize them to servers, so you get
- (single) client programs talking to replicated computations in the
- server. This is a 1-n pattern which you can easily solve in Isis,
- so the n-n issue isn't really seen...
-
- Comments? I think we should probably add n-n programming tools anyhow,
- but I am just curious to understand how commonly they would be needed!
-
- --
- Kenneth P. Birman E-mail: ken@cs.cornell.edu
- 4105 Upson Hall, Dept. of Computer Science TEL: 607 255-9199 (office)
- Cornell University Ithaca, NY 14853 (USA) FAX: 607 255-4428
-