NANOG (Bob Metcalfe) wrote:
     Perhaps I am confusing terms here. How can it be a fact that
     "store-and-forward delays are a mere fraction of wire propagation delays?"
     I don't think so. Check me on this:
     Packets travel over wires at large fractions of the speed of light, but
     then sadly at each hop they must be received, checked, routed, and then
     queued for forwarding. Do I have that right?
     Forget checking, routing, and queueing (ha!), and you get, I think, that
     store and forward delay is roughly proportional to the number of hops times
     packet length divided by circuit speed (N*P/C).
     For 10 hops of a thousand bit packet at Ethernet speed, that would be 1 ms,
     or a couple hundred miles of prop delay. Check me on this, one of us might
     be off by several orders of magnitude.


Using a real in use backbone of one of the mayor service providers,
I find that a DS3 between silicon valley to Chicago has a 44 msec
latency going through 4 hops. That's about the speed of light in
fiber for the 5000 mile roundtrip ICMP ping packets.

Using ATM will reduce the router latency. I estimate that with TCP/IP
over ATM over SONET OC-3c the latency will be reduced from 44 msec
to 40 msec, only a rather small improvement. The bandwidth used on the
fiber wont matter much. With OC-12c I would still expect 40 msec or so
since the speed of light in fiber is the limiting factor.



After all that, I knew IP/ATM had to show up somewhere! :wink:


Obviously, we need to find some faster light.