Wireless communication technology has gained widespread acceptance in recent years. Therefore, issues related to performance of transport protocols (such as TCP) over wireless links are of interest. The focus of this proposal is on TCP-unware techniques to improve TCP performance over paths that include wireless links. The path from a TCP sender to a TCP receiver usually includes several intermediate nodes which may drop TCP packets if congestion occurs. TCP makes the implicit assumption that all packet losses are due to congestion. Since wireless links are prone to transmission errors, this assumption is not accurate for TCP over wireless links. Nevertheless, TCP reduces its congestion window when packet losses due to transmission errors occur. This phenomenon is known to result in poor throughput for TCP over wireless links. Existing approaches for improving TCP performance over wireless links require intermediate nodes to take TCP-speciffic actions { these approaches are said to be TCP-aware. For instance, one such protocol requires that an intermediate node interpret TCP packets and drop some duplicate TCP acknowledgements when appropriate. Such protocols are often not useful in practice. For instance, when encryption is used for security, intermediate nodes may not be able to decipher contents of TCP packets. Therefore, TCP-aware protocols cannot be used. The objective of this project is to investigate TCP-unaware mechanisms that do not require any inter- mediate node to take TCP-speciffic actions. The following issues will be investigated:
- Design of TCP-unaware link layers that help improve TCP performance. The researchers will also consider other issues related to link layer design, including interaction of link layer with Random Early Detection (RED), providing diffierent reliability levels for TCP data and acknowledgements, and scheduling policies for link level retransmissions.
- Sender-based heuristics for distinguishing between congestion losses and wireless losses: Using such a heuristic, the sender can determine whether a packet was lost due to congestion or due to wireless transmission errors. Knowing the cause of a packet loss, the sender can take appropriate action for each type of packet loss. The impact of such heuristics on TCP performance will be evaluated.
- Receiver-based heuristics for distinguishing between congestion losses and wireless losses: When the receiver is closer to a wireless link, it may be in a better position to diagnose whether a packet is lost due to wireless transmission errors or not.
- Evaluation of a TCP-unaware approach that attempts to imitate a TCP-aware link layer mechanism. The proposed techniques will be evaluated using simulations as well as experimentation on a wireless testbed built with AT&T WaveLan technology. Educational Impact: Results from the research proposed here will be published in technical conferences and journals. The results will also be incorporated into graduate courses on networking and mobile computing. The software developed for the project will be used to design experiments for these courses, and disseminated for use by instructors in other universities. Students working on this project will gain expertise in an important technology area.
