The Next Generation Internet (NGI) poses scalability challenges to the efficient operation of the transport protocol (TCP). In particular, as the product (bandwidth x delay) grows, the congestion window required to .fill the pipe. becomes quite large, especially on cross-country links. One well known problem in TCP in this scenario is the selection of the .initial window.. More recently, new challenges have emerged because of the increasing popularity of nomadic access to the Internet via wireless links (e.g., wireless LANs, satellite links, UMTS, etc). The data rates on wireless links have been constantly on the rise, approaching the 50Mbps on 802.11a wireless LANs and thus providing an effective extension of backbone services to mobile users. However, wireless links tend to introduce random packet errors and loss that are not correlated to congestion. This creates problems to conventional TCP protocols (e.g., TCP New Reno and TCP SACK), which interpret any loss as a buffer overflow (i.e., as a symptom of congestion) and thus reduce the congestion window unnecessarily with consequent loss in performance. The drop in performance is proportional to the (bandwidth x length) product of the connection and can be quite significant in the high bandwidth NGI environment, especially on cross country paths including .last hop wireless LANs, UMTS links, or high bandwidth satellites. Several approaches to enhance TCP congestion control over high bandwidth wireless links have been reported in the literature (e.g., TCP Peach, TCP Westwood (TCPW)). Some of these enhancements have been quite successful. For example, TCPW, a TCP variant that uses bottleneck bandwidth share estimation. to adjust the congestion window upon loss, has shown scalable properties and good link utilization in large leaky pipes., (i.e. large bandwidth delay product, and non negligible random packet loss). This proposal is about carrying out a systematic, experimental investigation of performance of TCP over wired/wireless paths. This investigation will include the comparison of various TCP enhancements proposed so far in the literature. It will consider a representative set of experimental environments and application scenarios. The proposed project is in part motivated by our recent positive experience with TCPW Internet experiments of large file transfers over lossy paths. In this project the researchers will broaden the scope to include a vast gamut of TCP wireless enhancement techniques. The researchers will identify the pros and cons of each scheme, characterize the traffic/network environment for which it is best suited, and, more generally, develop models that relate wireless media characteristics, TCP congestion control parameters and performance results. In summary, given: (1) the increasing importance of nomadic computing and wireless access to the high speed wired Internet; (2) the performance degradations observed in conventional TCP protocols over wireless path, and (3) the encouraging improvements offered by modified, wireless versions of TCP (which yet retain the basic end to end paradigm), the researchers believe this a very appropriate time to engage in a systematic, experimentally based evaluation of wireless TCP protocols by a team that includes protocol developers, applications developers and network measurement experts.
