home *** CD-ROM | disk | FTP | other *** search
- Newsgroups: comp.dcom.cell-relay
- Path: sparky!uunet!elroy.jpl.nasa.gov!ames!pacbell.com!unet!trappist!earlf
- From: earlf@trappist.net.com (Earl Ferguson Adv Dev)
- Subject: Re: Path Building
- Message-ID: <1992Dec24.045318.18879@unet.net.com>
- Sender: news@unet.net.com
- Nntp-Posting-Host: trappist
- Organization: Network Equipment Technologies
- References: <1992Dec17.211811.6553@iscnvx.lmsc.lockheed.com>
- Date: Thu, 24 Dec 1992 04:53:18 GMT
- Lines: 48
-
- In article <1992Dec17.211811.6553@iscnvx.lmsc.lockheed.com> myoung@force.ssd.lmsc.lockheed.com writes:
- >OK, I have a problem here:
- >
- > I am a migrating process, and am currently sitting at memory
- >associated with a file containing objects of a certain class.
- >I have arrived here via a series of exploratory hops from data node
- >to data node, moving at each step when the local data met some
- >condition (I am a query moving through data).
- >
- > Now, having found the data I desire, I want to link this data directly
- >back to the location in memory where I started, thus building a memory
- >to memory VCI to satisfy my query. ...
- >
- > Problem: I want to traverse back over my path,
- Sounds like token ring source routing philosophy. Each exploratory hop
- adds its identification so that the return path can be examined. The
- return link is merely a traverse of the established path in reverse.
-
- >closing out any
- >loops appearing along the way back, possibly performing some local
- >optimization between hops looking for shorter paths. This backward
- >path I traverse, takes me back through switches, processors, and files
- >contained at processors, updating the virtual path along the way.
- >
- > Don't tell me that a hidden hand in the network management will do this
- >for me, as that just begs the question.
- Optimizing the return path requires some knowledge of the topology
- traversed beyond that that will be found in merely recording the forward
- path. Thus access to the topology data base (be it via network
- management or some generaly accessible topology server) and the path
- cost information (and desired COS/QOS, etc.) is all required.
-
- This doesn't sound like a networking issue as much as it is an
- application level issue, since there are factors that go well beyond the
- network layer that may need to be considered (e.g. if the data is
- replicated for security/backup, there are probably 2 or more possible
- results to the search, resulting in the need to further qualify which
- will be used - shortest, fastest, most reliable - and possibly
- coordinated with the results of previous searches and even possibly
- future searches [or simultaneous]. The decision perhaps needs to be made
- at the source based on the results of all other searches that should be
- coordinated in order to reduce queueing and synchronizing
- requirements.).
-
- ---
- Earl Ferguson
- earlf@net.com
- My personal opinions and NOT those of my employer.
-