Network Traac Characterization Using Token Bucket Model

| This paper investigated the problem of deriving token bucket parameters from the observed traac patterns of computer network ows in two cases. In the basic case, we identiied the set of token bucket parameters for any given ow so that all the packets in the ow will be delivered immediately without incurring any delay or loss. Then, we extended the basic case by adding a queue to the token bucket model to temporarily store packets that cannot be delivered right away until enough tokens get accumulated in the bucket. The queue has the eeect of smoothing the traac so that the resulting token bucket parameters are less demanding. It also makes the derived token bucket parameters less uctuated over time. Relationship among the queue size and token bucket parameters for a given ow are analyzed rigorously. Queuing delay for each packet in the ow is calculated and used to describe the adjusted post-queuing traac pattern. Simple and eecient algorithms for the derivation of measurement-based traac speciication (MBTS) are presented, along with empirical results. This MBTS technology alleviates the need for users to explicitly characterize the traac a priori in order to reserve network resource in an integrated services network. I. Introduction A. Conventions and problems Computer networks of integrated services, which provide multiple-level and dynamically selectable qualities of service (QoS) for diierent classes of traac, rely on sophisticated network resource reservation and management technologies. To fullll the service guarantee in an integrated services network 3], network traac needs to be characterized a priori in terms of some deterministic or stochastic traac models. This traac characterization provides a bounding envelope on the traac pattern that can be injected by the source to the network, and needs to be communicated to the network for resource reservation and admission control purposes. This paper focuses on the token bucket model (see 13, 9]) as adopted by the Internet Engineering Task Force (IETF) Integrated Services (IntServ) working group. One of the main challenges for such traac characterization is to derive token