home *** CD-ROM | disk | FTP | other *** search
- Newsgroups: comp.dcom.cell-relay
- Path: sparky!uunet!stanford.edu!kronos.arc.nasa.gov!iscnvx!news
- From: bantha.decnet.lockheed.com!young
- Subject: Re: Another Problem
- Message-ID: <1992Aug31.160529.8597@iscnvx.lmsc.lockheed.com>
- Reply-To: young@bantha.decnet.lockheed.com
- Sender: news@iscnvx.lmsc.lockheed.com (News)
- Organization: LMSC, Sunnyvale, California
- References: <1992Aug30.164226.2031@iscnvx.lmsc.lockheed.com>,<p8bpe4o@rhyolite.wpd.sgi.com>
- Date: Mon, 31 Aug 92 16:05:29 GMT
- Lines: 45
-
- In article <p8bpe4o@rhyolite.wpd.sgi.com>, vjs@rhyolite.wpd.sgi.com (Vernon Schryver) writes:
- >In article <1992Aug30.164226.2031@iscnvx.lmsc.lockheed.com>, bantha.decnet.lockheed.com!young writes:
- >> This newsgroup continues to debate the interface between TCP/IP routers
- >> and ATM. Must be an important application.
- >>
- >> Now here is another problem which ATM might solve.
- >>
- >> As I look around this company I see hundreds of little cubic foot computers
- >> on each desktop, exactly alike, distributed geographically. Some companies
- >> literally have thousands of these things.
- >>
- >> Imagine that an account executive wants to read into his spread sheet the
- >> entire daily summary of mail orders spread over a number of distributed
- >> offices. Thus, he wants to extract a particular file located in each of
- >> a large set of desktop machines.
- >>
- >
- >Are you sure that ATM is an obvious part of a solution to this kind of
- >problem?
- >
- >Wouldn't a naive ATM based solution involve setting up virtual circuits
- >to each of the target machines, some fixed number at a time, probably
- >one, with the consequent terrible response times? And wouldn't a naive
- >ATM based system involve a lot of manual configuring?
- >
- >Doesn't this sound more like an application for a multicast based
- >"distributed" application? Perhaps some kind of multicast remote
- >procedure call? Using UDP/IP multicast?
- >
- >That is not to say that the layers below UDP/IP might not involve one
- >or more wide area or local ATM circuits.
- >
- >
- >Vernon Schryver, vjs@sgi.com
- Yes, Yes the solution needs some kind of exploratory, application specific,
- protocol which can search out and find, then bind those desktops in
- a spanning tree, and then have the capacity to extract file data from
- each node, all in a parallel fashion.
-
- Does ATM help? Yes, if the vpi/vci are not naive. The intelligent
- messaging system performs best when it has a virtual "graph machine" below
- it on which it can build its tracks, tear them down, tunnel about etc.
- The problem is not that ATM virtual paths are naive, the problem is that
- the ATM architecture does not postulate a simple "bios" for path building
- and network exploration.
-