Performance of TCP traffic on HSDPA

  • Gutiérrez, Pablo Jose Ameigeiras, (Project Participant)
  • E. Mogensen, Preben (Project Participant)

Project Details

Description

The characteristics of the HSDPA transport channel (HS-DSCH) have been tailored to increase the packet data performance and deliver high user data rates while reaching significant capacity gain over the DCH based system. The time-shared nature of HS-DSCH is particularly beneficial to cope with the sources of channelization code inefficiency such as the inherent burstiness of the TCP?s slow start or the DCH inactivity timer. Hence, the interaction between the TCP flow control mechanisms and the particular characteristics of the HS-DSCH is to be addressed in this investigation. The result have shown that, in HSDPA, there exists a strong interaction between end user thoughput, system throughput, and packet queuing delay. The flow control functionality of the TCP protocol heavily participates in the aforementioned interaction and could easily influence the optimum operation point. If high user throughputs are to be provided (by lowering the offered cell load operating point), special attention is to be paid to the potential sources of end user throughput limitation introduced by TCP (such as the slow start, potential TCP segment losses, receiver buffer capabilities, etc). For the Proportional Fair scheduler, it has been shown that the decrease of the offered load below 1Mbps yields a median user throughput no larger than 420 kbps (in Pedestrian A at 3km/h, for the Proportional Fair scheduler, with a packet call of 100 kbytes and a round trip time rtt of 300 milliseconds). For heavy system loads, the cell throughput increases accordingly up to the maximum cell capacity, although the increased number of users limits the proportion of time every user holds the time-shared channel, and hence, restricts the achievable user throughput beyond the limits imposed by TCP. It is very relevant to note that the traffic burstiness caused by TCP does not influence the system capacity under such a heavy load. Moreover, under such conditions, the Node B queuing exhibits an exponential like growth with the load causing significant packet queuing delay at the Node B. (Pablo Ameigeiras, Preben E. Mogensen; Jeroen Wigard, Nokia)
StatusFinished
Effective start/end date31/12/200331/12/2003