CS268 Reading
Review
A Performance Comparison of Multi-Hop Wireless Ad Hoc Network
Routing Protocols
Josh Broch, David A. Maltz, David B. Johnson, Yih-Chun Hu and Jorjeta Jetcheva
Review by Feng Zhou
2/10/2003
The problem: Constructing a simulator for multi-hop wireless ad hoc
networks and using it to compare performance of different routing protocols. The
routing protocols are often too complex for analysis. So simulation is probably
the most feasible way to study their performance characteristics.
Key points:
- The implementation details of the simulation module provide useful
information about how wireless networks work. The collision detection mechanism
is very different from wired networks. And the "capture" phenomenon does not
exist in wired networks. Regarding the MAC layer, DCF of 802.11 uses explicit
RTS/CTS exchange to reserve the wireless channel before sending data. Link-
level ACK of every unicast message and retransmission are implemented to cope
with the problem of much higher random error rate than wired networks.
- Although a lot of effort is put into implement and optimize the four routing
protocols, and the simulation is well carried out, the scenarios used to study
their performance seem insufficient. One problem is the lack of TCP results.
Because TCP is used for almost all applications requiring reliable transmission,
actual application performance will directly depends on TCP performance. As
widely known, TCP performance is sensitive to random packet losses not caused by
congestion. Thus given that these routing protocols expose very different
packet loss patterns, studying how TCP perform on them will be interesting.
- Another problem is the use of long-lasting CBR flows. Although this models
applications like telephony and file transferring well, it does not model well
short-term flow pattern originating and targeting differnt hosts which are
exposed by applications such as HTTP and instant messaging. I expect these
routing protocols to display very different behavior under this flow pattern.
For instance, one difference will be that the short-term flow pattern is likely
to cause more routing overhead for protocols that use on-demand route discovery,
e.g. DSR. And this will in turn translates to much longer packet delivery
latency, which is an important metric not measured in this paper.