Research On QoS-Based Scheduling Algorithms Over An EPFTS Platform

With proliferation of network applications and drastic increase of multimedia traffic, Internet has been facing with four major challenges, i.e. high-speed switching, Quality of Service (QoS), security, and mobility. After unsuccessful attempts through enhancement over existing networks, the academy have realized the crux of NGI (Next Generation Internet) problem lies in the network architecture. The general background of this thesis is the research on an NGI architecture called SUPA (Single-layer User-data switching Platform Architecture) at SC-Netcom Lab (Sichuan Network Communication Technology Key Laboratory), and EPFTS (Ethernet-oriented Physical Frame Timeslot Switching) - the key technology to support SUPA.With the out-of-band signaling concept, SUPA seperates the User-data-switching platform (U-platform) from that for control and management information (S&M-platforms). By use of EPFTS, SUPA simplifies the typical three-layer User-data platform into single-layer structure, which benefits to improve switching efficiency and provide QoS guarantee. The primary goal at the first stage for SUPANET (network which support SUPA) is to focus on a high-speed switching substrate with QoS provisioning. The QoS provisioning framework in SUPANET involves QoS mechanisms both in S&M-platform and in U-platform. This dissertation is dedicated to realize QoS provisioning in U-platform with an emphasis on scheduling mechanisms.Traffic bursting is an important factor to cause congestion and performance degradation in traditional networks. In order to provide QoS guaranteed service in SUPANET, this dissertation tries to resolve end-to-end QoS provisioning problem by focusing on rate-controlling embed scheduling mechanisms to harness quantifiable QoS-parameters such as delay and delay jitters. As shown by the simulation results, the author's work is on the right track.With feasible implementation and scalability as two prerequisite conditions, this dissertation takes the design on rate-controlling based service strategy in EPFTS switching platform as a starting point, then carries out research on the rate-controlling embed scheduling mechanism base on single node, at last study on the end-to-end QoS provision solution through flow rate limitation between adjacent nodes in a multiple nodes interconnected network.The main work and contributions are as follows:First, grounded on the features of EPFTS, a new service strategy called STRC (Surplus Timeslot based Rate Controlling) is proposed. STRC utilizes the reserved timeslots of each flow as a measure to adjust the flow forwarding rate. In addition, two scheduling principles based on STRC - MASF (Mast Available Surplus timeslot First) and STRR (Surplus Timeslot Round-Robin) - are proposed. Analysis and simulation results show the effectiveness of the two principles to limit the burstiness of output traffic in addition to guarantee bandwidth.Second, limited to a single node, the scheduling mechanisms based on STRC for typical scalable switches are studied. For IQ (Input-Queued) switch consisting of crossbar and virtual output queuing, a packet scheduling strategy called TRWFS (Timeslot Reservation Weighted Fair Scheduling) and its implementation algorithms are proposed. For CICQ (Combined Input and Crosspoint Queued) switch, two scheduling mechanisms - "MASF/RR" and "STRR/RR" - are proposed. It is shown that these scheduling mechanisms are all effective in improving switch throughput and smoothing output traffic, at the cost of a modest delay increase.Third, a new rate-control scheduling mechanism called TRSFS (Smoothed Fair Scheduling based on Timeslot Reservation) is proposed. Extensive analysis and simulation show that TRSFS is Guaranteed Rate server, and TRSFS is easy to be implemented and capable of output burstiness limitation which helps to realize QoS provision in downstream node. TRSFS will lay the foundation for the end-to-end QoS provisioning to traffic flows in SUPANET.Fourth, with the scalability supportable, a solution called PPFAS (Port-Pair Fair Aggregation and Scheduling) is proposed to provide end-to-end QoS guarantee. By considering all micro-flows transferring through the same port-pair as one flow, PPFAS solves the scalable problem in backbone network. Moreover, by utilizing the CIOQ (Combined Input and Output Queuing) mechanism and TRSFS as the building block, PPFAS realizes a full distributed CIOQ switch. Extensive simulation show that at the cost of increasing traffic delay modestly, PPFAS solution features following benefits: (1) the interference impact between flows with different transfer path is weakened; (2) the end-to-end delay jitter is very small; (3) switches could reach 100% throughput under appropriate control of traffic load on switches.This dissertation show that though traffic is more delayed, rate-controlling based scheduling mechanism can bring more benefits of QoS performance. Research results in this dissertation are practical to be implemented in SUPANET. Especially the distributed CIOQ switch in PPFAS solution is viable to realize the end-to-end QoS provision in SUPANET.