| File system and RAW device are is widely used in key cluster system. These technologies are exposuring their disadvantages due to customer have higher and higher requirement about system availability.For file system, there are two problems. The first one is low redundancy, it can only be accessed by one server at any time. Although we may use high availability software to take over resources, the data is still not available during take over. Secondly, there are performance problem because others can't access the file system except the first server.RAW device is more available. Also, it has better performance because it can be concurrently accessed. However, there are some disadvantages too. Firstly, if RAW device is changed by any server, all other servers must synchronize these information offline. Secondly, we need to design RAW device carefully in order to avoid I/O hotspot. Also, after each time disk is added in RAW device, we need to re-design it to make sure keeping good performance. It is not convenient for management and availability.This paper first introduces some methodology about parallel file system. It can be concurrently accessed by more then one server. This overcomes above disadvantages of file system and RAW device and takes all advantages of them. Secondly, introduces the implementation process for cluster topology of new dispatching system, analyses availability of every module. Thirdly, summary a formula and parameters for high availability. Fourthly, compares the availability difference in details base on testing data, validates the effective performance of system availability. |
PDF Full Text Request