Internet Quality Of Service And Congestion Control

Congestion control mechanism of Internet and quality of service are two interrelated research subjects. For TCP/IP networks, packet drops is the most important method to send congestion notification to the source. When congestion occurs the router will drop the arrival packets and once the TCP source receives the notification it will back off the transmission rate to mitigate the congestion. But with the fast growth of unresponsive flows, such as multimedia flows, these two assumptions cannot come into existence for unresponsive flow doesn't adopt congestion control mechanisms, and therefore they will not back off the transmission rate when packet dropped. Due to these issues only using TCP congestion control mechanisms cannot assure the robustness of Internet any longer, instead, the router should actively participate in the network congestion control for the most effective detection of congestion can occur in the router itself. To address these problems IETF advocates deploying active queue management mechanism. AQM made routers have the ability to decide when and how many packets to be dropped thus stabilize the queue size effectively and avoid the problems of lock-out along with full queues. The essence of AQM is to detect the incipient congestion before queue overflow. When congestion has been detected, packets would be dropped or be marked (ECN) with a probability to send congestion notification to the sources.RED is the earliest and most popular AQM algorithm. The basic idea of RED is to monitor the average queue size to detect network congestion. When congestion occurs, RED randomly select sources to notify congestion before queue overflow and once TCP sources receive the notification it will reduce the congestion window to decrease the packet sending rate, and therefore, mitigate the network congestion. Because Internet is based on the bandwidth statistical multiplexing principle so there are lots of active flows competing limited bandwidth resource in a link. When network congestion occurs AQM must send congestion notification to adequate sources to reduce the traffic load and avoid queue overflow. Furthermore, AQM also need to avoid sending congestion notification to too many sources and consequently decreasing the utilization of bottleneck link. Therefore the sending rate of congestion notification should relate to the traffic load but RED doesn't take this factor into consideration thereby resulting in queue size oscillate which arose queuing delay and jitter. As an improvement, MRED (Modified RED) proposes an auto configure mechanism based on the changes of traffic load. When the traffic load becomes heavy MRED increases the value of maxp, and on the other hand when the traffic load becomes light MRED decreases the value of maxp, and therefore MRED can effectively adjust the congestion notification sending rate and stabilize the queue size.The Differentiated Services architecture, proposed by the IETF DiffServ Working Group has recently become the preferred service model for addressing QoS issues in IP networks because of its simplicity and scalability. Except the traditional Best Effort forwarding, the current DiffServ model also defines other two forwarding mechanisms: Expedited Forwarding and Assured Forwarding. The former forwarding mechanism provides a deterministic QoS and the latter provides statistical QoS. Accordingly, DiffServ model supports two services called Premium Services (PS) and Assured Services (AS). DiffServ is realized by marking each packet in its DSCP (Differentiated Services CodePoint) that determines the particular forwarding treatment. The Premium Service provides a guaranteed bandwidth, low delay, low jitter and low loss service that shows the same characteristics asa 'virtual leased line', and it's the best service in DiffServ networks. On the other hand, the Assured Service is intended to assure that a minimum throughput be provided even during periods of congestion. It permits to use additional available bandwidth when the network load is low. The basis...