Scheduling algorithms for high-speed switches
Department of Electrical and Computer Engineering
Doctor of Philosophy
virtual output queued (VOQ) switching architecture
quality of service (QoS)
earliest due date first matching (EDDFM) algorithm
The virtual output queued (VOQ) switching architecture was adopted for high speed switch implementation owing to its scalability and high throughput. An ideal VOQ algorithm should provide Quality of Service (QoS) with low complexity. However, none of the existing algorithms can meet these requirements.
Several algorithms for VOQ switches are introduced in this dissertation in order to improve upon existing algorithms in terms of implementation or QoS features. Initially, the earliest due date first matching (EDDFM) algorithm, which is stable for both uniform and non-uniform traffic patterns, is proposed. EDDFM has lower probability of cell overdue than other existing maximum weight matching algorithms. Then, the shadow departure time algorithm (SDTA) and iterative SDTA (ISDTA) are introduced. The QoS features of SDTA and ISDTA are better than other existing algorithms with the same computational complexity. Simulations show that the performance of a VOQ switch using ISDTA with a speedup of 1.5 is similar to that of an output queued (OQ) switch in terms of cell delay and throughput. Later, the enhanced Birkhoff-von Neumann decomposition (EBVND) algorithm based on the Birkhoff-von Neumann decomposition (BVND) algorithm, which can provide rate and cell delay guarantees, is introduced. Theoretical analysis shows that the performance of EBVND is better than BVND in terms of throughput and cell delay. Finally, the maximum credit first (MCF), the Enhanced MCF (EMCF), and the iterative MCF (IMCF) algorithms are presented. These new algorithms have the similar performance as BNVD, yet are easier to implement in practice.
njit-etd2001-074 (87 pages ~ 6,047 KB pdf)
Please complete this Feedback Form to inform us about your experience using this website. It will assist us in better serving your information needs in the future. Thank You!
Created October 04, 2002