Assume you wish to transfer an \(n\)-byte file along a path composed of the
source, destination, seven point-to-point links, and five switches. Suppose
each link has a propagation delay of \(2 \mathrm{~ms}\), bandwidth of \(4
\mathrm{Mbps}\), and that the switches support both circuit and packet
switching. Thus you can either break the file up into 1-KB packets, or set up
a circuit through the switches and send the file as one contiguous bit stream.
Suppose that packets have 24 bytes of packet header information and 1000 bytes
of payload, that store-and-forward packet processing at each switch incurs a 1
-ms delay after the packet has been completely received, that packets may be
sent continuously without waiting for acknowledgments, and that circuit setup
requires a 1-KB message to make one round-trip on the path incurring a 1-ms
delay at each switch after the message has been completely received. Assume
switches introduce no delay to data traversing a circuit. You may also assume
that file size is a multiple of 1000 bytes.
(a) For what file size \(n\) bytes is the total number of bytes sent across the
network less for circuits than for packets?
(b) For what file size \(n\) bytes is the total latency incurred before the
entire file arrives at the destination less for circuits than for packets?
(c) How sensitive are these results to the number of switches along the path?
To the bandwidth of the links? To the ratio of packet size to packet header
size?
(d) How accurate do you think this model of the relative merits of circuits
and packets is? Does it ignore important considerations that discredit one or
the other approach? If so, what are they?
