Study On Fairness Of Packet Marking And Queue Scheduling Strategies In DiffServ Networks

I. IntroductionDifferentiated Service Architecture(DiffServ) is presented by Internet Engineering Task Force (IETF) in order to provide QoS in the Internet. DiffServ has recently become a preferred method and a point of study for realizing QoS, because it has excellent extendability and robustness and it can be easily actualized.To realize the traffic control of IP QoS, DiffServ needs to use many kinds of controling mechanism, including the packet marking and queue scheduling strategies.At the boundary nodes in DiffServ, the AS flows should be regulated and put into different drop levels according to a certain standard of IN/ OUT. This can be carried out with the packet marker. The existing traditional marking algorithm can't adapt the impact of traffic features on the fairness in DiffServ networks.The queue scheduling is to allocate the bandwidth. It means choosing the packet to send from the waiting queues according to a certain rule. So it makes all traffic streams share the bandwidth of the output link by the established discipline. The traditional round robin based scheduling algorithm can't adapt the impact of packet size on the fairness.II. The Fairness Problems of the Assignment of Bandwidth in DiffServ(1) Because of TCP's itself congestion control mechanism, when TCP flows coexist with UDP flows, UDP flows will occupy more bandwidth than TCP flows.(2) Formula (3.7) presents that the excess bandwidth which is assigned to aggregates is the inverse ratio with CIR. So it is unfairness to the users who pay for the bigger CIR.(3) In DiffServ, the assignment of the bandwidth among the TCP flows is influenced by RTT and the packet sizes. The flow which has the shorter RTT or the bigger packet size will be assigned more bandwidth.III. Impovement of the Packet Marking AlgorithmIn this paper, we focus on solving the fairness problem by improving the marking algorithm. Our design aims at developing a marking system that is sensitive to the changing of the parameters such as RTT instead of measuring them directly. If a flow enjoys the network resources fairly in proportion, the ratio of its throughput to the total throughput should accord the ratio of its CIR to the total CIR. So in this paper we improve TSWTCM marking algorithm and design Dynamic Threshold Three Color Marking (DTTCM) algorithm.The design of DTTCM algorithm:(1) We use a fairness factor fairness = (thput / thputtotal )/(cir / cirtotal ) to apperceive the changing of the parameters. Simultaneity, we set the threshold minth = cir×(cir / cir max/ fairness)^1/2, max th= cir×(1+1/fairness)and the parameter max p = cir /[cir max×( 1+ fairness )]. Thereinto, The avgrate is the packet's average speed which is estimated. The thputtotal is the bandwidth of the bottleneck link. The cir is the CIR of the arriving packet. The cirtotal is the sum of all flows'CIR. The cirmax is the maximum CIR among all flows.(2) When avgrate≤cir, the probability of green is P_green=1. When cir≤avgrate≤maxth, P_green = minth/avgrate, the probability of yellow P_ywllow divides into [(cir)/(avgrate)- P_green] and [1 - [(avgrate-cir)/(maxth-cir)]²×(1-maxp)]×[1-(cir)/(avgrate)]. The preceding part is to guarantee the flow to achieve its CIR. And the latter part is to compete the excess bandwidth. When avgrate≥maxth, Pgreen = minth/avgrate, P_yellow = (cir)/(avgrate)-P_green. So it can give the flow some punishment to maintain other flows'interests. Under these three conditions, the probability of red P_red = 1-P_green-P_yellow.The simulation indicated that, comparing with TSWTCM algorithm, DTTCM has better performance in the assignment of bandwidth among UDP and TCP, as well as the assignment of excess bandwidth among the aggregates. In aspect of the impact of packet size and RTT on the fairness, when the network provision level is low, DTTCM has better performance. But when the network provision level is high, DTTCM is not better than TSWTCM.IV. WFQ Queue Scheduling MechanismThe existing round robin based scheduling algorithm can't adapt the impact of packet size on the fairness. Generalized Processor Sharing (GPS) is one of an ideal mechanism that can resolve this kind of problem. This kind of algorithm supposes that the packet is infinite separable, and the router can serve all active queues simultaneity in proportion. So it can make all kinds of flows share the bandwidth fairly.GPS is only an ideal model, and it is unable to realize actually. The paper presents Weight Fair Queue (WFQ) algorithm which is based on Packet Generalized Processor Sharing (PGPS) to simulate GPS algorithm which is based on fluid-reference-system. In the GPS system, the rate ri which is assigned by weight will be changed by the change of the number of back logged queues B j. So WFQ uses a virtual time measurement to schedule every packet. The algorithm maintains a virtual start time and a virtual finish time for every back logged queue. When there are events, WFQ will update the virtual finish time for every back logged queue by updating the virtual time function V(t ) and then chose the packet whose virtual finish time is the least to send. Virtual time reflects the resource of time that every back logged queue provides to all service. It follows the real time when the server works. Since the rate of the server is constant, virtual time also reflects the data quantity which the server provides to all back logged queues.The simulation experiment indicate that when the packets are different size, the link can also assigne the bandwidth fairly by WFQ scheduling strategy. V. Conclusion(1) DTTCM has better performance in the assignment of bandwidth among UDP and TCP, as well as the assignment of excess bandwidth among the aggregates.(2) In aspect of the impact of packet size and RTT on the fairness, when the network provision level is low, DTTCM has better performance. But when the network provision level is high, DTTCM is not better than TSWTCM. It is because when the TCP flows can't be easily achieve their CIR, the DTTCM marker can't improve the fairness by marking more yellow and red packets to the flows which has short RTT or big packet size.(3) Comparing with the round robin based scheduling algorithm, WFQ algorithm can preferably simulate the ideal GPS system. So it can make the link assign the bandwidth fairly, even the size of the packets are different.