The growth of higher bandwidth networks and powerful new workstations has enabled the development of many new multimedia applications. These applications involve the combined use of different media such as voice, video and text. Supporting these applications requires the resolution of a number of issues both within the network and at the end-hosts that originate the applications. In this dissertation we consider real-time scheduling and protocol problems that arise in this arena. In the realm of applications operating under hard time-constraints, we address both network-node and workstation scheduling. In applications involving the use of video or a combination of voice and video, there can be the need to schedule classes of traffic with differing importance but identical deadlines at a network-node. We examine scheduling algorithms that can be used in the situation of two classes of traffic. We demonstrate using probabilistic arguments that a new "balancing" discipline can provide better loss performance for both classes than more traditional schemes in some operating regions. When delay loss probabilities are low, strict priority scheduling is shown to be a reasonable option. Turning our attention to the end-hosts, we consider the need to process multimedia objects such as voice packets in a periodic fashion within a workstation, and study two common scheduling algorithms--earliest deadline first (ED) and rate monotonic (RM). Recent studies have revealed the importance of the preemption behavior of the scheduler in determining overall processor performance. We prove that there are always fewer preemptions under ED than under RM and show through simulations of well-known task-sets that, on occasion, there can be over 20% more preemptions under RM than under ED. Apart from these scheduling problems, we also look at protocol issues that combine network and host considerations. For applications such as world-wide multimedia lectures, it becomes important to have error-recovery protocols that will scale well to hundreds of receivers. We perform throughput analyses and demonstrate why receiver-initiated protocols based on negative acknowledgements are to be preferred over sender-initiated protocols based on positive acknowledgments. We focus on host processing costs to illustrate this result.
| Identifer | oai:union.ndltd.org:UMASS/oai:scholarworks.umass.edu:dissertations-8958 |
| Date | 01 January 1994 |
| Creators | Pingali, Sridhar |
| Publisher | ScholarWorks@UMass Amherst |
| Source Sets | University of Massachusetts, Amherst |
| Language | English |
| Detected Language | English |
| Type | text |
| Source | Doctoral Dissertations Available from Proquest |
Page generated in 0.0014 seconds