Per-flow delay performance in traffic aggregates

Abstract

Abstract-Class-based traffic treatment frameworks such as Differentiated Service (DiffServ) have been proposed to resolve the poor scalability problem in the flow-based approach. Although the performance is differentiated in a class-based basis, the performance seen by individual flows in the same class may differ from that seen by the class and has not been well understood. We investigate this issue by simulation in a single node under FIFO, static priority, waiting time priority, and weighted fair queueing scheduling schemes. Our results indicate that such performance discrepancy occurs especially when flows joining the same class are heterogeneous, which is not uncommon considering that the same type of applications can generate traffic having very different statistical behaviors such as video traffic with different activity levels, or voice traffic with different compression schemes. We found that per-flow delay statistics, including the average and the 99 th percentile delay, can be very different from the corresponding class delay statistics, depending on flow burstiness, overall traffic load, as well as the queue discipline. We also propose a solution to reduce the mean delay variance experienced by flows in the same class. Symposium: Internet Performance I. INTRODUCTIONTo overcome poor scalability and high complexity in per-flow traffic handling, per-aggregate or class-based packet scheduling has been suggested (as in the DiffServ architecture [1]), where QoS differentiation is made on a per-class basis. However, the implicit assumption is that the per-flow performance and the aggregate/class performance is close, which may not hold under certain circumstances, as shown in this paper. Equating the per-flow QoS and the class performance when they in fact could disagree degrades the perceived application performance, or deteriorates the integrity of service differentiation. For instance, if the packet delay bound or mean delay of class B is supposed to be twice that of class A, it could happen that some flow in class A sees a delay bound 20% worse than the class while some in class B sees a delay bound 20% better than the class. This certainly is unfair to class A users since the ratio of the delays is now 1.33 instead of 2, as expected. This phenomenon calls for the attention to identify quantitative deviations between the perflow QoS and the class QoS so that applications can be made aware of what QoS levels to expect from the network.

Opportunistic link overbooking for resource efficiency under per-flow service guarantee

“…Example 1: If delay constraint → ∞, no waiting time constraint is imposed on the packets. We then have ( )| =∞ = 1, and from equation (17) and (2), → ; while in equation (16), → = 1 − . Thus, we can calculate packet loss rate from equation (15) as…”

“…Two problems still remain: 1) how to improve the efficiency of the dedicated link; 2) how to maintain the link's QoS capability; and few work has addressed the trade-off issues between them when developing per-flow QoS. Tremendous endeavors have focused on providing flow level service guarantee based on the Diffserv architecture (see, e.g., [14][15][16][17][18][19]), and much other work has put their emphasis on improving the efficiency and scalability by dynamic resource management (see, e.g., [6,[20][21][22]). Besides, seldom work has aimed at developing an integrated analytical framework on flow level, which is critical for seeking the balance between link efficiency and QoS guarantee to determine the proper link overbooking factor, while considering the different requirements of delay-sensitive and loss-sensitive flows.…”

Flow assignment method with traffic characteristics over multiple paths for reducing queuing delay

“…In distributing the traffic flows, however, inadequate traffic balancing into multiple paths significantly degrades queuing delay performance, especially when the path is highly utilized [5][6][7]. When different (bursty and nonbursty) types of flows are to be balanced over multiple paths, the non-bursty flows assigned to the same path as the bursty flows encounter the same queuing delay performance as the bursty ones.…”

Analysis of Individual Flows Performance for Delay Sensitive Applications

“…Results suggest that individual flow delay varies with respect to aggregate flow delay, according to bandwidth utilization level and the variation of the traffic burstiness level. A similar study realized by Siripongwutikorn and Banerjee has investigated individual flows delay embedded in a single traffic class, by considering the provisioning strategy based on aggregate traffic [13]. They considered several scheduling disciplines, such as FIFO and WFQ in their analysis, and the results indicate that traffic heterogeneity, network load and scheduling disciplines affect individual flows performance.…”