[ee122] Follow up on the Problem about SNW vs. Sliding Window

Vern Paxson vern at icir.org
Wed Oct 3 23:43:08 PDT 2007


> I've come across two definitions of bandwidth-delay product in my search for
> answers, one of which uses the on-way delay and another which uses the RTT
> delay.
> 
> I looked in the book and they use the <d sub prop> and is interpeted as one
> way delay. (HW1 [p.73 P.18])

This is a good point to clarify.

P18 in HW1 was emphasizing what it takes to fill a link, and in that case
(since no feedback from acks is needed to keep the process going [by sliding
the window]), the one-way delay suffices.

However, for sliding window, the delay waiting for the acks to arrive is
a critical element.  Thus, when considering transport performance, you
need to use the RTT rather than the one-way time.  (In addition, for a
precise calculation this RTT needs to account for both the propagation
time and the transmission time.  This latter will be different in the two
directions if the data being sent differs in size from the acknowledgments
being returned to slide the window.)  By the way, this case of considering
the RTT is the usual use of bandwidth-delay product.

		Vern


More information about the ee122 mailing list