High speed communication support for multimedia applications: tree network and reconfigurable transport protocol
Hung Khei Huang · 1993
The development of high-speed network is essential to adequately support an increasingly diverse distributed multimedia applications that are available. Many of those applications have communication requirements that extend beyond those found in traditional data applications. Both the underlying network and transport protocol infrastructure are very important factors that significantly affect the performance of high-speed networks. The underlying network infrastructure implements the lower-layer, link-to-link protocols primarily in hardware. The transport protocol infrastructure integrates higher-layer, end-to-end network protocols. In this dissertation, we address issues on both network and transport protocol infrastructures. First, as an underlying network infrastructure, we propose a new high speed network architecture called Collision Avoidance Multiple Broadcast (CAMB) Tree network. The network consists of collision avoidance switches organized in a tree topology. Each switch is an internal node in the tree. Stations are the leaves of the tree. The switches allow the implementation of random access protocols (simple and easy implementation) without its major disadvantage: the penalty of collisions among packets. Tree network solves the problem of packet collisions without incurring overheads (e.g., access coordination) introduced by controlled access protocols (e.g., FDDI, and DQDB). It combines the benefits of random access (low delay when traffic is light; simple, distributed, and robust protocols) with concurrency of transmission, excellent network utilization and suitability for the domain of high-speed optical networking. A design and implementation of the Tree network is presented. Performance results are obtained from the network prototype implemented, validating the concepts of the CAMB Tree Network. We also build a simulation model, based on the prototype. We compared the performance of the Tree Network with other network architectures (Ethernet, Token-Ring, FDDI), showing better performance results for the Tree Network. Simulations of large networks are also presented. Second, in the transport protocol infrastructure area, we propose a transport protocol reconfiguration scheme. Reconfiguration provides adaptivity to transport protocol, enabling it to adjust to the highly diverse and dynamic applications and network characteristics. We designed and implemented the reconfiguration scheme for one of the major transport protocol functions: error handling. Three error handling mechanisms are considered: (1) cumulative acknowledgement with go-back-n, (2) selective acknowledgment with selective repeat and (3) forward error correction. We empirically evaluate these mechanisms using several different applications (such as data, voice and video) that run in a simulated network environment. Through these experiments we determine (1) efficient combinations of application/mechanism/network parameters, (2) the application and network conditions under which mechanism switching reconfiguration is advantageous, and (3) the reconfiguration process overhead.