Simulation And Performance Analysis Of Scheduling Algorithms For EPFTS

The background of this thesis is the new network architecture called SUPA (Single physical layer User-data transfer & switching Platform Architecture) aiming at Next Generation Internet. To simplify the User-data transfer & switching platform into a single layer, a novel technique called EPFTS (Ethernet-oriented Physical Frame Timeslot Switching) is introduced. EPFTS combines the framing and switching functions of the data link layer with multiplexing function to a physical channel.The work presented in this thesis is QoS-guaranteed techniques pertinent to switch fabric and switching mechanisms of U-platform. It focuses on queuing issues and scheduling algorithms in EPFTS. The emphasis has been given to quota-constraint multi-priority queuing scheduling algorithms and efficient switching & forwarding core, which is purposed to reduce the complexity of traditional switch fabric and related scheduling algorithms.The paper first describes the basic principles of queuing scheduling under the input and output queuing, and analyzes their performance by simulations. Secondly, it introduces priority-based queuing scheduling algorithms in consideration of the services' QoS requirements. Aimed at the unfair shortcoming of strict priority-based queuing algorithms and the regulations of QoS Negotiation Protocol under the EPFTS environment, it promotes a quota-constraint multi-priority queuing scheduling algorithms (Quota-PQ), which can get better QoS guarantee for each class service. Under the different quotas of each class service, the thesis simulates its effects to the queuing algorithms. Furthermore, the thesis simulates the effects of self-similarity characteristic to queuing scheduling algorithms by modeling a self-similar source, which is synthesized by many sources and each satisfies Pareto distribution. At last, the thesis presents a new switch & forwarding core, which has many advantages with low complexity.The thesis's main work and contributions can be summarized as:(1) It describes the non- or prioritized queuing scheduling principles under the input and output queuing, and analyzes their performance by simulations;(2) It promotes a quota-constraint multi-priority queuing algorithms, which avoids the shortcoming of absolute priority-based scheduling algorithms and can get better QoS guarantee for each class service. Under the different quotas of each class service, the results of simulations indicate that the new algorithm is valuable for switch design.(3) Considering of self-similarity characteristic of service flow, the paper models the self-similar source and simulates its effects to scheduling algorithms, contrasting with the results under the Poisson traffic.(4) To reduce the complexity of N×N switch fabric and its scheduling algorithms, the thesis presents an efficient switching & forwarding core N-Input-Single-Output Multiplexed Express Forwarding Structure (NISO-MEFS). It reduces the complexity to N×1, and avoids the multiple match processes at the input and output ports in input queuing algorithms. Furthermore, its performance approaches to the one that is obtained under the output queuing, but it does not require so high to switch fabric as output queuing does.