Application Driven Evaluation, Design, and Implementation of Network Services and Resource Sharing Models

点击翻译按钮获取中文摘要

The focus of this proposal is on problems that arise in the context of QoS enabled networks and their use by applications. In particular, the main theme of the proposal, is the development of a greatersynergy between potential users of network QoS capabilities, i.e., applications, and the design and implementations of those QoS capabilities within the network. In order to accomplish such a goal, work will be carried out along the following three major axes:1. Use and selection of network QoS services and parameters by applications. The goal here is to identify performance parameters that are of most significance to applications, and translate that knowledge into {\emutility} curves that can then be used by both applications to select the appropriate network service, and by the network to make intelligent decisions on how to best handle an application's traffic in case of resources contention. This includes not only assessing sensitivity to traditional QoS parameters such as bandwidth, loss, delay, and jitter of different types of applications traffic, e.g., transactions, audio, video, etc., but also considering more complex scenarios involving applications with multiple traffic streams whichmight involve different (dynamic) resource sharing rules depending on the availability of network resources.2. Algorithms for performance monitoring, resource sharing, and dynamic resource allocation in support of both network and applications requirements. The goal here is to design and evaluate specific network levelmechanisms in support of applications requirements for dynamic sharing of resources and service monitoring. For example, access to a ``high quality'' service by certain traffic streams might be predicated on the presence or absence of other streams. Devising and assessing mechanisms allowing the specification and implementation of such functionality can greatly improve the ability of networks to deliver services that are useful to applications. The challenges are not only to design mechanisms capable of supporting applications requirements, but also to integrate them with existing network services and capabilities. For example, it may require coupling between services above and beyond what is currently supported, e.g., ensure small delay to an audio stream, but only as long as a given floor rate can be guaranteed to data transfers.3. Design and implementation of programmability in network devices in support of applications requirements and new service models. The goal of this last dimension of the proposal is to investigate the possibility of supporting applications needs and some of the above associated mechanisms, by leveraging some of the programmability that is being made available in a number of new network devices. The benefits of such an approach, if feasible, are that it can foster incremental deployment as well as facilitate the introduction of new capabilities if and when required by new applications. The challenges are to identify the appropriate trade-off between functionality and the implementation constraints that the technology imposes. The investigation of the above topics will be carried out through both experimental and analytical efforts. Experimentation will take place in the context of a newly established lab devoted to multimedia andnetworking, that includes end-systems as sources of application traffic, and networking equipment made available by several vendors. Furthermore, work on the design and implementation of new mechanismsto better support application service requirements, will be done in close collaboration with equipment vendors so as to facilitate their incorporation and testing on the available platforms. Finally, participation in the QBone Internet2 testbed will allow us to carry out end-to-end experiments that would not be feasible with only a local testbed. In particular, the QBone environment provides a representative sample of the type of network QoS services that will be initially available, and the experiments that will be carried out overit will offer the opportunity to investigate not only how applications should use them, but also to identify related problems that may arise in their initial deployment.