Scheduling Strategies of Video Cache in a Video Server

A Video-on-Demand system (VoD) [1] integrates the entertainment, telecommunication, and computer industries and provides electronic video rental services to geographically dispersed users from remote video servers on a broadband network. Several techniques have been proposed to increase the number of concurrent viewers of popular movies through better usage of resources. One technique uses the batching method in which one I/O stream is dedicated to service several viewers who have arrived within the same timing window [2]. Another technique for reducing the amount of I/O resource needed is the buffering method, where viewers can fetch the data blocks from the VoD system buffers in stead of the disk subsystem [3]. However, in addition to the normal playback, the VoD system is expected to provide VCR functions such as jump-forward, jump-backward, fast-forward, rewind, and pause. Although batching and buffering techniques can service requests for popular movies in large bulks, they do not work very well in the presence of interactive VCR functions. For staggered streams, Almeroth and Ammar [6] uses the set-top box buffer to provide limited interactive functions. Yu et al. [7] develops the look-ahead scheduling with set-aside buffer protocol, which attempts to take the advantage of batching but only supports the interactive operations of pause and resume. Another approach requires a new buffer space to handle user interactions for each user, who will hold onto this stream until disconnection [8]. This approach works only if very few users issue interactive operations. Otherwise, the system may start in a batch mode, but degrade to a nonsharing mode as more and more users split off into their own streams. Liao and Li [4] use Split and Merge (SAM) to take the share of a video stream transparent to users while they allow truly user interactivity. They use synchronization (synch) buffer, shared by all users, to reach truly interactive functions. However, this method spent too much buffer space while users become more and more. Park and Ryou [5] use the dynamic buffer to reduce server I/O and network bandwidth requirement, and the dynamic buffer can be supported by interactive VCR functions to all users. This method still needs more buffer space while the number of users becomes large. Thus, although video cache buffer plays a very important role in the video server, we found that the current solutions to solve interactive services-like VCR functions in the video server require the extra video cache buffer space. In this paper, we propose a new scheme that can support an interactive VCR for all users requesting the same video stream with batching and does not require more video cache space. In our scheme, we include a video cache manager and add the scheme of admission control to satisfy the real-time playback and disk bandwidth constraints in order to avoid the infinite waiting of users. Through our C-based simulation for the proposed algorithm of processing interactive requests, we have that the interactive requests issued by users do not require extra video cache buffer space. As number of users becomes large, the waiting probability is getting down oppositely. This is a primary benefit in the video server. So, it comes out to meet the real interactive requests and satisfy the sprit of the VoD with our proposed video cache system.