Multimedia Streaming and Proxy Caching

Abstract

The information technology revolution has led to the explosive growth of the Internet. There is an increased demand for real-time applications like audio/video streaming and videoconferencing. The increase in demand for real-time applications, coupled with the convergence of digital multimedia and the World Wide Web, has ultimately led to the growth of a new technology called Internet Multimedia. Multimedia transmission over the Internet takes the form of Streaming. Multimedia objects are delay sensitive and have timing constraints. Multimedia streaming is resource intensive and as demand for multimedia data over the Internet increases, the network will not be able to handle the traffic, leading to scalability and latency problems. An elegant way to tackle these problems is through the use of proxy caches. Caching video objects is difficult because video objects are typically large in size and cannot be stored in totality at the proxy cache. Video streaming takes place with the use of real-time protocols like RTSP and RTP and the proxy must be able to interact with the main server and client/s using these protocols. This thesis deals with the use of proxy servers that cache streaming video objects. We emphasize the need for an RTSP enabled proxy server that can communicate with a group of clients as well as with the main server using RTSP. The thesis describes a streaming architecture for layered video objects using a proxy. The thesis also uses the concept of layered caching, where the video is stored in the form of layers at the cache to increase cache utilization. Quality of the streams delivered from the cache is also an important aspect of proxy caching. Our thesis also proposes to measure the quality of streams cached at the proxy. Our thesis presents a framework for video caching. A method for the efficient use of reserved bandwidth is proposed. The bandwidth allocated using RSVP is used to increase the quality of cached streams. Quality enhancement is achieved with the help of an RTSP enabled proxy that exists between the main server and the client/s. A new weight assignment policy based on the playout time of a video object at the client is proposed. The popularity of the object is evaluated using the average playout time at the client across all sessions and weight is assigned based on the popularity of the object. Other weight assignment schemes are based on frequency of access, recency of access and size of object and are compared with the new scheme. A new metric to measure the quality of the cached streams is proposed. The quality metric proposes to use information about the number of layers of each object present in the cache, the total number of layers of the object as well as the relative perceptual significance of each layer, to measure the quality of the object. The thesis also brings out a comparative study of various replacement policies. We study the performances of the existing replacement policies and the proposed new replacement policy in video caching experiments. The performance of fine-grained replacement technique for layered video objects is evaluated. The performances of fine-grained replacement technique and atomic replacement technique for layered video objects are compared.