No abstract

Abstract-It is a widely held belief that packet reordering in the Internet is a pathological behavior, or more precisely, that it is an uncommon behavior caused by incorrect or malfunctioning network components. Some studies of Internet traffic have reported seeing occasional packet reordering events and ascribed these events to "route fluttering", router "pauses" or simply to broken equipment. We have found, however, that parallelism in Internet components and links is causing packet reordering under normal operation and that the incidence of packet reordering appears to be substantially higher than previousl reported. More importantly, we observe that in the presence of massive packet reordering Transmission Control Protocol (TCP) performance can be profoundly effected. Perhaps the most disturbing observation about TCP's behavior is that large scale and largely random reordering on the part of the network can lead to selfreinforcingly poor performance from TCP.

In this paper, we propose to use the idealized Hierarchical Generalized Processor Sharing (H-GPS) model to simultaneously support guaranteed real-time, rate-adaptive besteffort, and controlled link-sharing services. We design Hierarchical Packet Fair Queueing (H-PFQ) algorithms to approximate H-GPS by using one-level variable-rate PFQ servers as basic building blocks. By computing the system virtual time and per packet virtual start/finish times in unit of bits instead of seconds, most of the PFQ algorithms in the literature can be properly defined as variable-rate servers. We develop techniques to analyze delay and fairness properties of variable-rate and hierarchical PFQ servers. We demonstrate that in order to provide tight delay bounds with an H-PFQ server, it is essential for the one-level PFQ servers to have small Worst-case Fair Indices (WFI). We propose a newp PFQ algorithm called WF 2 Q+ that is the first to have all the following three properties: 1) providing the tightest delay bound among all PFQ algorithms; 2) having the smallest WFI among all PFQ algorithms; and 3) having a relatively low asymptotic complexity of O(log N). Simulation results are presented to evaluate the delay and link-sharing properties of H-WF 2 Q+, H-WFQ, H-SFQ, and H-SCFQ.

Hierarchical Packet Fair Queueing (H-PFQ) algorithms have the potential to simultaneously support guaranteed realtime service, rate-adaptive best-eort, and controlled linksharing service. In this paper, we design practical H-PFQ algorithms by using one-level Packet Fair Queueing (PFQ) servers as basic building blocks, and develop techniques to analyze delay and fairness properties of the resulted H-PFQ servers. We demonstrate that, in order to provide tight delay bounds in a H-PFQ server, it is essential for the onelevel PFQ servers to have small Worst-case Fair Indices (WFI). We propose a new one-level PFQ algorithm called WF 2 Q+ that is the rst to have all the following three properties: (a) providing the tightest delay bound among all PFQ algorithms; (b) having the smallest WFI among all PFQ algorithms; and (c) having a relatively low implementation complexity of O(log N). We show that practical H-PFQ algorithms can be implemented by using WF 2 Q+ as the basic building block and prove that the resulting H-WF 2 Q+ algorithms provide similar delay bounds and bandwidth distribution as those provided by a H-GPS server. Simulation experiments are presented to evaluate the proposed algorithm.

We consider the definition of the Expedited Forwarding Per-Hop Behaviour (EF PHB) as given in RFC 2598 [1], and its impact on worst case end-to-end delay jitter. On one hand, the definition in RFC 2598 can be used to predict extremely low end-to-end delay jitter, independent of the network scale. On the other hand, we find that the worst case delay jitter can be made arbitrarily large, if we allow networks to become arbitrarily large; this is in contradiction with the previous statement. We analyze where the contradiction originates, and find the explanation. It resides in the fact that the definition in RFC 2598 is not easily implementable in schedulers we know of, mainly because it is not formal enough, and also because it does not contain an error term. We propose a new definition for the EF PHB, called "Packet Scale Rate Guarantee", which preserves the spirit of RFC 2598, while allowing a number of reasonable implementations, and has very useful properties for per-node and end-to-end network engineering. We show that this definition is stronger than the rate-latency service curve guarantee. Then we propose some proven bounds on delay jitter for networks implementing this new definition, both in cases without loss and with loss.