Deadline-based scheduling in support of real-time data delivery
Computer Networks: The International Journal of Computer and Telecommunications Networking
A delay pricing scheme for real-time delivery in deadline-based networks
WINE'05 Proceedings of the First international conference on Internet and Network Economics
| Hi-index | 0.00 |
Transport of real-time data over packet networks is becoming increasingly important. To ensure timely delivery of real-time data, quality of service (QoS) support at the transport network is required. Traditionally, IP-based networks provide best-effort service; their delay performance is good when the traffic is light. However, they do not scale well when the level of traffic increases. To date, most strategies to manage network resources in order to provide QoS support are based on the notion of a data flow; each flow specifies its bandwidth and performance requirements at the transport network, and QoS support is provided on a per-flow basis. For certain types of real-time data transfer, such as the delivery of a small document, it might be difficult to specify a flow with given bandwidth requirements, because data will not be presented to the network for a significant period of time. We have developed a novel approach to managing network resources in order to support real-time applications. The key element of this approach is the use of a (size, deadline) pair to characterize each real-time application data unit (ADU). The bandwidth requirement of the ADU is implicitly specified by the ratio: size/(deadline - current time). The ADU deadline can also be mapped into packet deadlines, which can be carried by packets and used by routers for channel scheduling. An efficient and effective deadline-based scheduling algorithm is developed. In this algorithm, packets that have smaller ratios of (time until deadline) over (number of hops remaining) are given higher priority. This algorithm has constant time complexity, and achieves performance comparable to that of the best known algorithm in terms of the percentage of ADU's that are delivered on time. Our algorithm is also effective in large networks that have multiple autonomous systems, and in multi-service configurations. When the network traffic is heavy, congestion may occur and channel scheduling alone may not be able to prevent degradation in system performance. Mechanisms that can alleviate the load on the network are needed. We have designed two new application-layer ADU admission control algorithms. Both algorithms use feedback information from the receiver to control the rate at which new ADU's are accepted. One of them also uses the bandwidth requirements of an ADU in making admission control decisions. Simulation results comparing the performance of these two algorithms are presented.