home *** CD-ROM | disk | FTP | other *** search
- Newsgroups: comp.dcom.cell-relay
- Path: sparky!uunet!mcsun!sun4nl!research.ptt.nl!walvdrk
- From: walvdrk@research.ptt.nl (KEES VAN DER WAL)
- Subject: Re: > Congestion Avoidance
- Message-ID: <1992Jul28.104127.1@research.ptt.nl>
- Sender: usenet@spider.research.ptt.nl (USEnet News)
- Nntp-Posting-Host: dnlts0
- Organization: PTT Research, The Netherlands
- References: <1992Jul27.165554.8612@sics.se>
- Date: Tue, 28 Jul 1992 09:41:27 GMT
- Lines: 36
-
- In article <1992Jul27.165554.8612@sics.se>, craig@sics.se (Craig Partridge)
- writes:
-
- > There's plenty of work going on to try to avoid having ATM use the horribly
- > wasteful peak bandwidth approach. (Why is it wasteful? Well, because
- > the peak bandwidth typically isn't being used, but by having chopped the
- > full circuit bandwidth into connections, you've made sharing of the unused
- > bandwidth harder. It is usually better to completely share the entire
- > bandwidth of the link, than to break it up into smaller fixed bandwidth
- > channels).
-
- You've got to make a choice: either you stress the "guarantee" for throughput
- and a low loss figure _or_ a high degree of multiplexing gain. A combination of
- both is harder to obtain.
-
- You can have _both_ approaches mixed in a single ATM network.
-
-
- > One key problem is that data communications traffic is typically not
- > nicely IID and thus M/*/n queueing models are of little help. (The
- > Sunshine switch designers reported that when they tried a more realistic
- > model of data traffic they found the buffering requirements were an
- > order of magnitude larger than M/*/n models predicted). One can rapidly
- > find oneself analyzing queueing theory problems that cause experts to
- > shudder.
-
- What "more realistic" model did they use?
-
-
- > However, there are some nice ideas. Some of the token bucket schemes
- > (a variant of leaky bucket) have been shown to give delay and bandwidth
- > guarantees over cell networks. See Parekh's MIT dissertation for one.
-
- Looks interesting; is this kind of literature available somewhere?
-
- Regards, <kees>
-