The Study On The Improvement Of The Load Balance Algorithm For FastDFS Distributed File System

With the growing popularity of the rapid development of information technology and the Internet, a single server storage mode is difficult to supply the data storage and management in business units. Along with the growing demand for distributed storage, the Applications of Distributed File System is more and more widely, which makes the storage servers increasingly higher requirements.In this thesis, FastDFS is used as the research object. According to the file system architecture and design ideas, the problem of load inclination in the file system is pointed out, and analyzes the reason for the problem. By the combination with the load inclination problem and the classic load balancing algorithms, a more suitable load balancing algorithm for FastDFS is presented. It can not only guarantee the system's ability of load distribution, but also can avoid load imbalance when the system is linear expansion and overcome the existing load balancing problem. As an innovation of improving the load balance algorithm of FastDFS, this thesis completed the following tasks:First of all, it analyzes the architecture and design of FastDFS, such as lightweight, the grouping way and the peer structure. And then make a deep explanation of data's redundant backup and files' synchronization. During the analysis, the problem of load inclination comes out based on the FastDFS's load balancing strategy when the linear expansion, and the key factor of the problem is deeply analyzed.Secondly, the dynamic strategy of weighted round robin load balancing algorithm based on service types is pointed out by combining classic load balancing advantage and application scenarios. And the algorithm description and algorithm flowchart are both displayed. This algorithm collects every server's load status information regularly, and calculates the corresponding load value, pick the servers by the certain range of load value. According to the remaining space of each server, the weight values are calculated in the picked group, which will be used for load distribution by the weighted round robin algorithm. The algorithm fully considers the storage server's work load in real-time and the remaining storage space, and groups the servers range of load value to calculate the weight of each server by its' remaining space, and then distributes the load for each server combining the weighted round robin algorithm with its' advantage. It can make servers dynamic load balance using smaller cost, so that each servers can play a maximum performance.Finally, the improved algorithm is programmed in the implementation process for certain search algorithm optimized and the way of grouping management is implemented according to the load values and the real-time load of servers are recorded by using shared memory. Use virtual machine clusters to set up FastDFS clusters, simulate a small FastDFS distributed file systems, and then verify the validity and superiority of the improved algorithm in experiment.