Study On Video-on-demand Enabled Scalable Distributed Streaming Delivery Approaches

With the popularization of broadband and the development of computer technologies, supplying streaming services simultaneously to a mass of users in Internet is becoming possible, which is also being payed more and more attentions. In live streaming field, peer-to-peer based approaches come to a great success, and the commercial application is speeding up. However, in video-on-demand field, relating researches make progress laggardly because of the asynchronism of users'request to program. As a result, many technique problems are still left to be solved.How to get rid of the contradiction between the limited streaming server serving capability and the great requirement of network resources caused by parallel request of users is the core problem of video-on-demand enabled scalable streaming transmission. Cluster based approaches are limited in scalability because of its centralized data transmission manner. Content delivery network based approaches are very costly, thus are impossible to be applied extensively. Most of existing infrastructure nonsupport of support IP multicast, which makes multicast approaches hard to be deployed. Peer-to-peer based approaches won't cause aforementioned problems, and is known as"the more users, the better service". What's more, the number of users this kind of approaches support is independent of the capability of streaming servers, which makes them being seemed as the most promising approaches.New characters of peer-to-peer based approaches of video-on-demand are analyzed in the paper. Many existing solutions, such as buffer management, partner finding and multi-send data transmission.is also studied. It is found that the application of diskette space based buffer management policy would be held back by copyright leakiness and the probable harmfulness of user's diskettes, although the policy may enable peers to store and share a large amount of program slices. On the other hand, presented memory space based buffer management policy often can't make a good balance between the utility of buffer space and the timely arriving rate of program slices. Centralized partner searching policy is efficient, but bad in scalability because of bottle-neck problem of index server. Media distribute graph based partner policy is inefficient and brittle. Hash table based partner finding policy is efficient and scalable, but still feazed by many problems. Existing multi-sender data transmission algorithms are predominant in some aspects, but faultful in other aspects for instance, they may be not very suitable to the character of buffer policy of video-on-demand, or may be too complicated to implement, or relate too much on streaming server, etc.On the basis of the study of existing video-on-demand techniques and systems, an innovative peer-to-peer based approach is presented in the paper. To expand the capability of servers, muli-sever t. overlay topology is adopted. By using index-serial table based partner fast searching algorithm, the delay of booting is limited, and VCR operations could be surported very well.The buffer management policy of the approach mainly contains two aspects, buffer mamngement policy of servers and buffer mamngement policy of peers. In terms of buffer mamngement policy of servers, a popularity-random hybrid policy is presented, which enable the freely expansion of the capability of streaming servers. In terms of buffer mamngement policy of peers, the policy of this paper makes a good balance between the utility of buffer and the continuous playing back of programs by adding a pre-fetch buffer segment to the prefixal buffer.To adapt to the change of overlay topology caused by the join and leave of peers, Gossip, which is applied extensively in live steaming and some video-on-demand systems, such as Gridcast, is improved in the paper. The main feature of the improved protocol is that the spread scale of membership messages is confined to a necessary range, which greatly improves the capability of peers to adjust their local topology according to the change of the overlay. Such improvement is very propitious to keep the program that the user demand playing back steadily, and lessen the request of streaming servers.A smallest-dead-time-first based multi-sender data transmission algorithm is also presented. Multi-sender data transmission algorithm count for much to the scalability and Qos of the system. Considering there are much less eligible partner for peers in video-on-demand systems than in live streaming systems, partners are divided into two types: peer partners and server partners. By using the capability of peer partners greedily, the pressure imposed on streaming servers is lessened. By downloading the most needed slices first, the playback continuity of programs is assured.At the end of the paper, the improvement of the protocol and algorithm in the approach is measured and validated by simulation.