home *** CD-ROM | disk | FTP | other *** search
- Newsgroups: comp.dcom.cell-relay
- Path: sparky!uunet!munnari.oz.au!yoyo.aarnet.edu.au!news.adelaide.edu.au!augean.eleceng.adelaide.edu.AU!ksarkies
- From: ksarkies@augean.eleceng.adelaide.edu.AU (Ken W. Sarkies)
- Subject: Re: ATM "style" (was: Lockheed Switch)
- Message-ID: <1992Nov10.065059.3466@augean.eleceng.adelaide.edu.AU>
- Keywords: ATM switching traffic
- Organization: Electrical and Electronic Eng., University of Adelaide
- References: <ru9l3ps@sgi.sgi.com>
- Date: Tue, 10 Nov 1992 06:50:59 GMT
- Lines: 72
-
- In article <ru9l3ps@sgi.sgi.com> rpw3@rigden.wpd.sgi.com (Rob Warnock) writes:
- >myoung@force.ssd.lmsc.lockheed.com writes:
- >+---------------
- >| I had a request to post info on the Lockheed switch... We have proposed
- >| a bufferless control structure... A modified deflection routing scheme...
- >| at each network switch, a data packet must bid for an output link... by
- >| providing an ordered list of links for each packet... Conflicting bids
- >| for the same output port are nondeterministically solved.... implemented
- >| using technologies that are scalable both in speed and network configuration.
- >+---------------
- >
- >In further conversations with one of the researchers referred to me by Matt,
- >I found that this switching scheme is really quite clever. I wouldn't be at
- >all surprised if it could be used to build inexpensive multi-gigabit LANs.
- >
- >However, the "deflection routing" architecture inherently re-orders packets
- >(or frames, cells, whatever), very severely when put under a heavy load.
- >(After all, it handles congestion by routing packets "somewhere else" for
- >a while.) Using it in an ATM environment would be a real challenge. Using
- >it directly with unmodified AAL1, AAL2, or AAL5 would be impossible.
- >
- >This points up one of the really fallouts that I predict is about to
- >occur from our industry-wide rush towards ATM (and yes, I include myself
- >in that rush). There are many kinds of potentially useful network and
- >switch architectures which are going to be pushed aside because they are
- >incompatible with the ATM "style" (small cells, ordered delivery, bandwidth
- >declared at call setup, even the very requirement of "call setup").
- >
- ..... other stuff deleted
-
- >
-
- In this department we have been working on ATM switch architectures for a
- few years, and are now looking at performance aspects of various general
- structures. A few general considerations which need to be kept in mind are
-
- 1. The BISDN is supposed to be a high speed network, so we do not want
- to be encumbered by the protocols. They should be simple and efficient.
- For that reason it is reasonable to look for switches which do not break
- up cell ordering.
-
- 2. Switches must work at least as fast as the optical fibre links to which
- they are connected. This is not likely to be possible electronically
- without a great deal of hardware and parallelism. They should not be more
- complex than is necessary.
-
- 3. Under normal conditions, switches will not be overloaded longer than
- the buffers can handle. Rarely, brief overloads may occur on selected outputs.
- A switch which recirculates blocked cells would be very useful, but not
- necessarily essential. However cell reordering can be done at the switch
- outputs, with little problem (probably in output buffers), to satisfy the
- standards.
-
- 4. Under overload conditions (allowed by the traffic control in the network),
- switches can be overloaded for longer periods. In such cases it is expected
- that cells will be lost, in which case recirculation, deflection routing
- etc. will not help. It is important only that in the switch, innocent traffic
- is protected. Switches which have recirculation or other rerouting of overload
- traffic may have the potential for the overload traffic to cause cell loss
- on streams which are not passing to overloaded outputs. That in my opinion
- is where the real problem with such architectures lies.
-
- I believe that ultimately the simpler switches will turn out to be preferred
- on the basis that they can be produced cheaper and faster and more flexible
- than the more
- complex architectures. Note that this is not a criticism of the Lockheed
- architecture, but a general comment.
-
- --
- Ken Sarkies
-
- No disclaimer. My boss takes all responsibility for my big mouth.
-