Performance of TCP traffic over WCDMA

Packet data services are the distinctive characteristic of Third Generation wireless networks. In many cases, the information provided by these services will be accessible on the Internet. According to measurements carried out by Miller in April 1998, 95% of the overall data traffic that crossed a middle Internet backbone was TCP traffic. The majority of Internet?s traffic is implemented over TCP because this is a stable, mature, and probably the most thoroughly tested transport protocol of its kind. Most of the interactive and background services to be addressed by WCDMA networks are expected to belong to this vast group of applications employing a TCP/IP platform, which defies UMTS to properly handle the TCP/IP traffic. The properties and characteristics of the incoming data to the WCDMA radio network completely depend on the protocol stack of every service. This project has analyzed the performance of TCP over WCDMA due to the wide application of this protocol and the strong impact of its flow control on the behaviour of the traffic, and ultimately on the performance of WCDMA. On a general perspective, it can be concluded that there exists a significant performance difference between low and high bit rate channels for realistic round trip times (115-450 ms). The higher bit rates allocated by the Packet Scheduler to the user, the bigger is the potential risk of the congestion window not reaching the bandwidth delay product. This risk can lead to the user throughput being degraded below the allocated channel bit rate, as well as to low traffic activity factors. The low traffic activity is also an undesired effect because it enhances the variance of the overall Node B transmission power (i.e. risk of network instability), and it can even lead to channelization code shortages especially in microcell environments. The results have shown that, for small file sizes, such as web pages, the initial slow start can harm the user throughput of users allocated high channel bit rates. A user that is allocated a 256 kbps channel with a round trip time (rtt) of 140ms, experiences 25% throughput degradation with a 50 kbytes file, whereas a 150 kbytes file reduces the degradation down to 11%. Users assigned lower bit rates channels, such as a 64 kbps experience a higher file download delay, but hardly suffer any degradation from the initial slow start for round trip times (rtt) of 150 or 300 ms. Regarding the interaction between the RLC and the TCP layer, it has been shown that the more reliable is the RLC layer, the more benefits are obtained in terms of user throughput because the end-to-end TCP recovery mechanisms have negative long-term effects on the congestion window. This conclusion is obtained even considering RLC retransmission delays up to 180 ms, because the round up of the RTO to one second and the reset of the TCP retransmission timer whenever a new acknowledgement is received, offer considerable protection against spurious time outs caused by a large number of RLC retransmissions. (Pablo Ameigeiras, Preben E. Mogensen; Jeroen Wigard, Nokia)