Dynamic congestion control and error recovery over a heterogeneous Internet
Upamanyu Madhow · 2002
In the heterogeneous Internet of the future, end-to-end transport protocols for error recovery and congestion control must be able to handle large bandwidth delay products, random loss (i.e., loss due to phenomena which occur on a time scale faster than the feedback delay), and possibly asymmetry between the bottleneck link capacities on the data and acknowledgement (ACK) paths. Our objective in this paper is to provide a flavor of transport protocol design and analysis in this scenario by illustrating two key points: (a) Although rigorous analysis for transport protocols over heterogeneous networks is difficult, approximate analysis can be used to obtain rules of thumb that are invaluable for designing network level controls for supporting transport protocols. This is illustrated by showing how our recently developed analytical understanding of TCP/IP, the popular Internet transport protocol, leads to network control mechanisms for hiding the time variations and losses of a wireless link from TCP, or for dealing with the effect of asymmetry. (b) While it is not clear that the feedback structure required to apply traditional control schemes is available at the transport layer, application of systems level thinking in protocol design can lead to significant performance gains over TCP/IP. We illustrate this using our recent work on a transport protocol based on a novel ACK format that conveys information sufficient for error recovery and congestion control even with an ACK path of poor quality.