Cache management policies for streaming proxies : the cases of fixed and variable video sizes and collaboration among proxies

Abstract

The easy access on the Internet, the exponential growth of the Internet traffic and the ever enlarging sizes of files which are transferred over it, often lead to network congestion and significant delays in many applications that require timely data transfer. The most important part of data transfer over the Internet corresponds to media streaming applications which require a large portion of the available bandwidth, while at the same time there is the need of achieving low delays, because the delays in these applications are very annoying to the users and some of them may not be willing to experience them. In order to improve the services provided to the users of these applications, the techniques of video caching and of collaborative caching are used. Video caching embodies proxy servers on the network in order to keep frequently accessed data at locations close to the clients. Collaborative caching uses group of proxies which can collaborate with each other, and achieve better performance than with independent standalone proxies. This Thesis proposes new efficient algorithms for video caching which manage the cache of the proxy servers for video transfer in real time and compares them to existing algorithms. Also, a new effective collaborative scenario is proposed and evaluated. In the first two parts of the Thesis the performance of three existing efficient algorithms in the literature, the Number of Common Clients - Size (NCCS), the Least Recently and Least Frequently Used (LRLFU) and the Least Recently Used (LRU) is evaluated and compared for the cases of fixed and variable video sizes. In each of the first two parts of the Thesis a new cache management algorithm is also designed and proposed, this new algorithm consists of a combination of existing efficient algorithms. In the first part of the Thesis the proposed new algorithm (referred to as the LRLFU_Aggressive) does not perform much better than the existing algorithms, but in the second part the proposed new algorithm (referred to as the LRLFU_Modified) significantly improves the performance of the system in comparison to the existing algorithms. In the cache replacement algorithms we examine a value defined for each video, namely the caching value, which is based on the popularity of the videos, in order to retain in the cache the most popular videos. For each requested video which does not exist into the cache, other videos have to be removed (the ones with the lower caching values) in order for the new video to be cached. The performance metrics we use to evaluate and compare the algorithms in our system are the Video Hit Ratio and the Fraction of the Delayed Starts. In the third part of the Thesis a collaborative scheme of two client communities each associated with one proxy server, is defined and studied. While most of the studies in the area of collaborative proxies use the same cache management algorithm in each proxy and assume that each client community communicates via the corresponding proxy with the same content server, in our collaborative system we have used a different cache management algorithm in each of the two proxies and we assumed that there are two content servers each one associated with one of the client communities. Furthermore, we assumed that each of the client communities communicates via the corresponding proxy only with its corresponding content server. Also, we assumed that there is an overlap between the contents of the two content servers. Each of the two different cache management algorithms used in the collaborative scheme caches different videos in the corresponding proxy, and this helps the overall system performance. From the results of our simulation study we derive important conclusions for the performance of the examined collaborative scheme.