Multicast Congestion Control in UMTS
Congestion control is a policy that adapts the source transmission rate according to the network congestion. In Internet Protocol (IP) multicast, the User Datagram Protocol (UDP) is used for the transport layer. This protocol does not implement any congestion control. Instead, the Transmission Control Protocol (TCP) adapts its transmission rate according to network congestion. This means that the coexistence of multicast traffic and TCP traffic may lead to unfair use of network resources. In order to prevent this situation, the deployment of multicast congestion control is indispensable. This kind of congestion control is well-known as TCP-friendliness.
The adoption of a multicast congestion control in cellular networks poses an additional set of challenges which are related to the existence of radio links and mobile terminals. All the algorithms for congestion control treat the packet loss as a manifestation of network congestion. This assumption may not apply to networks with radio links, in which packet loss is often induced by reasons other than network congestion like noise or radio link error. As a consequence, the network reaction should not be a drastic reduction of the sender's transmission rate. Moreover, the mobile terminals' computing power cannot afford complicated statistics and traffic measurements. Consequently, such operations should not be executed on the mobile equipment.
LDST investigates the applicability of two well known multicast congestion control schemes over UMTS networks. We analyze the applicability of the TCP-Friendly Multicast Congestion Control (TFMCC) and the Pragmatic General Multicast Congestion Control (PGMCC) to UMTS networks. Both schemes belong to the class of single-rate congestion control schemes. In this class of schemes, the receiver with the worst congestion level is selected as the representative and the transmissions rate is adjusted accordingly. Such schemes are simple enough, so as to meet a prime objective for UMTS multicast services, that is scalability to applications with thousands of receivers.
TFMCC (TCP-Friendly Multicast Congestion Control)
TFMCC is a well known equation-based multicast congestion control mechanism which is fair towards competing TCP flows. It uses a control equation derived from the TCP equilibrium equation which relates the long term throughput to the loss and Round-Trip Time (RTT) (this scheme is called "equation-based" for this reason). This equation directly controls the sender's transmission rate. The loss event rate and the RTT are the parameters that define this target throughput. Each receiver calculates its target throughput and considers it as the acceptable sending rate from the sender to itself. TFMCC uses a feedback scheme which assures that the feedback of the receiver calculating the slowest transmission rate, always reaches the sender. This scheme is based on the concept of the Current Limiting Receiver (CLR). Moreover, the TFMCC design ensures that the sender gets feedback from the receivers experiencing the worst network conditions without being overwhelmed by feedback (feedback implosion is suppressed).
PGMCC (Pragmatic General Multicast Congestion Control)
PGMCC is a TCP-friendly scheme in which the receiver reports are sent to the sender in the form of Negative Acknowledgments (NAKs). This functionality helps the sender to continuously monitor the status of the receivers' group. The receiver with the worst throughput according to the control scheme is the group's representative, the acker. The receiver reports embedded into the NAKs help the sender to select and to track changes of the acker. A window-based congestion control scheme similar to TCP congestion control is run between the sender and the acker, which has the responsibility to send positive ACKs for each data packet. The window-based control used in PGMCC is partly different from TCP congestion control. The differences are: the use of distinct windows for rate and for reliability/flow control, the retransmission behavior, the use of sender-based RTT measurements for selecting the representative and the ACK clocking scheme when the representative switches over.
The degradation of the radio channels in the UTRAN causes malfunctions in the legacy TFMCC and PGMCC schemes. The innovation of LDST work stems from the fact that the original schemes are partly modified and extended in order to support the particularities of the UTRAN. Our proposals introduce minor modifications in the UMTS architecture. Additionally, complicated operations like statistics and traffic measurements are avoided to be performed on mobile equipment. Therefore, our schemes take into account the limited computing power of the mobile terminals. Last but not least, it was our motivation to study the modified TFMCC and PGMCC schemes in a comparative way.