Request Scheduling in Advance Reservations with Partial Uncertainty

• Main Authors: Kai-Hui Lee, 李開暉
• Other Authors: Yie-Tarng Chen
• Format: Others
• Language: en_US
• Published: 2002
• Online Access: http://ndltd.ncl.edu.tw/handle/67344443958616709744

博士 === 國立臺灣科技大學 === 電子工程系 === 90 === Resource reservation in advance is a potential approach to provide Quality of Service (QoS) for next-generation network. However, reserving resources in advance introduce resource fragmentation and scalability problem in resource management. Resource fragmentation not only increases blocking probability but also reduces resource utilization. On the other hand, Scalability problem becomes a bottleneck for supporting QoS on backbone network. Hence, the major purpose of this thesis is to address these two issues. In this thesis, we propose a flexible reservation model with deferred acknowledgement to support request scheduling. In this model, the starting time of each book-ahead request is partial uncertainty, i.e. the starting time in request specification is an interval rather than an exact time. Hence, the acknowledgement of the starting time for book-ahead requests is deferred until the end of a reservation cycle instead of immediate decision and acknowledgement. Based on the flexible reservation model, two request-scheduling algorithms are proposed to tackle the resource fragmentation problem. First, we propose an immediate request-scheduling algorithm for a single-link environment. Whenever a new book-ahead request cannot be admitted, the immediate request scheduling is invoked to find sufficient resources to satisfy the new request. Since, the resource allocation in flexible intervals can be represented as a multistage digraph, optimal request scheduling is equivalent to find a shortest path in the digraph. Hence, we employ the dynamic programming approach to solve this problem. Second, we propose a deferred request-scheduling algorithm on a subnet. When a book-ahead request arrives, an admission control with oversubscribed model is employed to classify the book-ahead request into the admitted request, the pending request or the rejected request. At end of the reservation cycle, a deferred request scheduling is applied to find a new schedule for admitted requests and pending requests under admission constraints such that the overall resource utilization can be maximized. Finally, the admission decision for the pending requests and the starting time for admitted requests are announced. However, the request scheduling on a subnet is a combinatorial problem. We use divide-and-conquer approach to partition original scheduling problem into several small sub-problems in a small interval. Each sub-problem is equivalent to the 0/1 Multi-Dimensional Knapsack Problem (MDKP), which is a well-known NP-hard problem. Furthermore, a heuristic algorithm for the MDKP is proposed to speed up the scheduling process. Simulation results show that the proposed request-scheduling algorithms can notably improve the resource utilization and reduce the blocking probability. Scalability is the main problem of providing QoS guaranteed on the Internet. The advance reservation system has to maintain large amount of reservation information such that the problem becomes more critical. In this thesis, we propose a scalable architecture and a flow aggregation algorithm to attack the problem. However, impolitic flow aggregation can violate the original admission test. Therefore, the objective of the proposed algorithm is to reduce the number of aggregated flow without any lose in bandwidth requirement. Simulation result indicates that the number of aggregated flow can be reduced effectively even in the worst case. In addition, in the best case, the revenue gain in bandwidth approaches to the mathematical upper bound.