Copy Storage Strategy Streaming Media Server Cluster

In recent years, Internet based stream media applications such as video conference, IP phone, and Video on demand services have become more and more popular. However, due to limited network resources and too many client nodes, stream media servers are often under more loads than they could afford and thus could not provide high quality of service to their clients connected. Compared with traditional Web servers, requirements for stream media servers' performance optimization are far stronger than that for Web servers. As a result, it becomes critical to work out mechanisms to effectively optimize performance, reasonably allocate server resources and depressurize the stream media servers.In stream media server cluster, how to distribute media data and scheduling between data is paramount to the system's performance. Yet the system's performance is directly impacted by the replica placement algorithms, which is directly impacted by the density of the data requests, the popularity of media files, the connection session lengths, as well as media files' bit rates. The most direct phenomenon is the not-evenly-distributed feature of hot films will directly cause some servers to overload, but the other servers are far from full loaded at the same time. Nevertheless, the density of requests, the popularity of media, the session lengths and medias' bit rates are usually not a constant factor, which increased the difficulty to simulate the cluster's load.To solve the problem, this paper gives an Erlang-B formula based algorithm for replica placement on video on demand server cluster. This algorithm worked out a mechanism to media file replication under the four variables mentioned above. It firstly deems the whole user requests as a Poisson process, and utilizes Erlang-B formula to get minimum global request rejection rate. Afterwards, resources needed for each video to reach the global rejection rate is calculated and mapped to video copies on the servers in the cluster.Simulation experiment shows that this algorithm can lower the rejection rate of users' requests, and balance the user satisfaction on all files, and lowered storage space requirement.