This paper discusses the role of wireless link modeling in the design of transport protocols. Due to non-congestion-related packet losses and the latency associated with corrective link-layer mechanisms, wireless links can have a drastic effect on TCP transport performance. The authors present that a wireless link model should be simple but realistic - as long as the model simulates the net effect of a mechanism, it need not completely represent the actual mechanism. The authors describe the topology and traffic characteristics for cellular, WLAN and satellite wireless links. Transport protocols are evaluated using "goodput", the portion of delivered data that is useful, as well as the standard performance metrics: throughput, delay, fairness and dynamics. Having introduced the work, the authors present models for corruption, delay variation, reordering, channel allocation, bandwidth variation, asymmetry, buffering and handovers.
Finally, the authors comment on the interplay between transport protocols and wireless technologies. Characteristics that are not fundamental to transport protocols (or wireless technologies) should be modified to accommodate smooth interaction with wireless technologies (or transport protocols). Having read the previous paper on TCP performance over wireless links, this one was very appropriate and provided a more balanced approach to the unison of transport-layer protocol and link-layer technology. One thing I'm still confused about is the following statement: "Furthermore, because cellular and satellite links have high latency, connections traversing them can be penalized by the bias of TCP congestion control toward flows with high RTT (Round-Trip Time)". Assuming that high latency is related to high RTT, why does a congestion control mechanism biased towards connections with high RTT "penalize" such connections?
Subscribe to:
Post Comments (Atom)
No comments:
Post a Comment