home *** CD-ROM | disk | FTP | other *** search
- Newsgroups: comp.dcom.cell-relay
- Path: sparky!uunet!spool.mu.edu!uwm.edu!linac!att!att!allegra!hgs
- From: hgs@allegra.att.com (Henning G. Schulzrinne)
- Subject: Re: Bandwidth Control
- Message-ID: <1993Jan25.125059.20389@allegra.att.com>
- Organization: AT&T Bell Laboratories, Murray Hill, NJ
- References: <3402@ucl-cs.uucp>
- Date: Mon, 25 Jan 1993 12:50:59 GMT
- Lines: 45
-
- In article <3402@ucl-cs.uucp> J.Crowcroft@cs.ucl.ac.uk (Jon Crowcroft) writes:
- >
- >
- > kyma@shell.portal.com (Matt J Young) writes:
- > >+---------------
- > >| Experts analysis has shown that statistical multiplexing
- > >| over the WAN is simply not possible.
- > >+---------------
- >
- > Rob Warnock, MS-9U/510 rpw3@sgi.com replies:
- > >Then they need some new experts. At DCA, we were doing "statistical
- > >multiplexing over the WAN" with a cell-relay-like system in 1971, and
- > >people have been doing it ever since. It is *easier* to do in a WAN
- > >environment, since you are averaging more traffic sources.
- >
- >absolutely they need new experts - sounds like someone is quoting
- >professional litigation rytpe "experts":-)
- >
- >papers from bellcore and cabruidge computer labs report on as much as
- >6-10 fold increase in network utilisation for video traffic if you use stat
- >muxing rather than fixed peak bandwidth allocation
- >
- >i havnt seen any quality papers contradict this (based on garret's
- >video stats from domestic entertainment video and cambridges real
- >video conferecing stats wit hthe pandor system over CFRs (i.e. real
- >ATM style nets)....
- >
- >jon
-
- The paper by Reibman and Berger, Globecom '92, p. 314ff does a fairly
- in-depth comparison of CBR and VBR video (subject to leaky-bucket
- constraints). It shows that significant multiplexing gains are possible,
- but with leaky-bucket constraints and without knowing the source
- behavior in advance, it will be extremely difficult to get more than
- a factor of 2 if you want loss guarantees and if network buffers are
- small. Depending on your viewpoint, this means that either
- 1) leaky buckets and guarantees are a 'bad thing' if you want cheap
- video service
-
- 2) anything but peak rate allocation is going to be difficult, unless
- you can describe your source very accurately (e.g., for video-on-demand)
-
-
- Henning Schulzrinne
- hgs@research.att.com
-