Re: [RE] On worst-case latency for Ethernet networks and alternative shaping concept
At 16:40 28/09/2005, Gross, Kevin wrote:
[snip]
>The whole ATM experience is an interesting reference point. The objectives were technically and practically sound. They got off to a good start - Reducing all data streams to a series of 53 byte cells does make traffic engineering a tractable problem. ATM is no more complex than it needs to be. And yet the complexity turns out to be untenable and cost prohibitive in most non-telecom applications.
I agree with all that except for the penultimate sentence; ATM actually is quite a bit more complex than it needs to be. There's a whole lot of baggage which is a legacy of its history in telecoms, plus I personally think they only need two of the eight combinations of class of service and connection configuration (UBR unicast and CBR multicast -- so no VBR or ABR) and most switches don't implement CBR in the most efficient way.
I've just become aware of http://100x100network.org/ which has a lot of good stuff on this topic, for instance at the bottom of page C-8 of the document at http://100x100network.org/papers/100x100proposal.pdf pointing out that circuit switching is inherently cheaper and more scalable than packet switching. An ATM CBR service done properly is more like circuit switching.
John Grant
___ ___ ___ ___ ___ ___ ___ ___ ___
| || || || | | || || || || |
| N || i || n || e | | T || i || l || e || s |
|___||___||___||___| |___||___||___||___||___|
Nine Tiles Networks Ltd, Cambridge, England
+44 1223 862599 and +44 1223 511455