[e2e] TCP out of order

Fred Baker fred at cisco.com
Wed Jan 23 17:28:08 PST 2002


At 11:51 PM 1/22/2002, Michael Welzl wrote:
> > Note that you can get very consistent packet reordering even if the
> > speed-of-light delay difference is zero and the circuit bandwidths
> > are identical, if the packets being generated are sometimes different
> > sizes.  That is, if you send a big packet followed by a small packet, the
> > small packet will always catch up to and pass the big packet if they
> > traverse enough store-and-forward hops.
>
>Forgive my ignorance - but: why?

in addition to issues mentioned, chaos theory applies. If two packets 
arrive simultaneously at a router, cross a backplane, and go out the same 
interface, either one might go "next", and the second will wait for the first.

Further, the queue depth on any given interface is not zero. Normally, 
networks are well overprovisioned, so p/(1-p) comes up with a small number, 
but it is a random number averaging a certain value, it is not a certain value.


At 01:16 AM 1/23/2002, Michael Welzl wrote:
>Just out of curiosity: doesn't this pose a problem for the packet pair
>approach?

yes and no. PP generates an estimate of the link bandwidth available, and 
one would expect that it would normally be an estimate of the *available* 
bandwidth on a bottleneck link, since in a packet network any packet 
presumably has O(p/(1-p)) packets in queue on arrival of each of the 
packets in the pair. What it means is that there is some error in the 
estimate that wouldn't be there if everything were indeed deterministic.

But please, what does the word "estimate" mean? I think we already know 
that it is an initial approximation to a value that will vary over time.




More information about the end2end-interest mailing list