From: Michelle Liu (liujing@u.washington.edu)
Date: Tue Oct 26 2004 - 20:44:13 PDT
Review of "Supporting Real-Time Application in an Integrated Services Packet Networks: Architecture and Mechanism"
Jing Liu
This paper proposes an Integrated Service Packet Network (ISPN) architecture that supports two distinct kinds of real-time service: guaranteed service and predicted service.
The author characterizes network clients to two types: adaptive and tolerant or rigid and intolerant. With different service requirements, the scheduling algorithm, service interface to these two types of clients are different. With flow isolation scheme of weighted fair queueing (WFQ), guaranteed service could have a bounded worst-case delay. While for predicted service, WFQ is not the best choice considering utilization of network bandwidth and delay jitter. FIFO is preferred for sharing and allows a number of sources aggregating their traffic to obtain a lower overall delay jitter. A unified scheduling algorithm is presented to integrate both guaranteed service traffic and predicted service traffic. Time-stamp based WFQ scheme is used as a framework to provide both services. Within this framework, per-flow based scheduling is used to provide guaranteed service and strict priority classes and FIFO+ scheduling are used to provide predicted service. With regard to these two different real-time services, different service interface for each and admission control are also discussed in this paper.
The advantage of this paper is that the author proposes a model which is quite practical to implement QoS in networks. Although per-flow based WFQ scheduling is good to provide guaranteed service, the status maintaining and overhead in routers can not be ignored and also the scalability has to be concerned. Moreover, quite a lot of applications do not require such a restrictive guaranteed service. They are adaptive and tolerant or they require the network to provide predicted service. Thus, the integration of both services is necessary.
However, it seems today the Internet is good enough for most of applications. The link bandwidth has been increased dramatically. When we design QoS protocols, we have to consider the evolution of network hardware and whether the protocol is really practical and necessary.
This archive was generated by hypermail 2.1.6 : Tue Oct 26 2004 - 20:44:30 PDT