Research And Implementation Of Load Balancing Technology For Distributed Block Storage System

In the trend of cloud computing and big data, distributed block storage system become increasingly important because of its unique advantages. Load balancing is an important characteristic of a distributed block storage system, and also is one of the hotspots of distributed block storage research. Based on the existing distributed load balancing algorithm, study new more effective load balancing technology to ensure the stability of the system performance is of great significance.First,based on consistent hashing(CHT) algorithm with virtual nodes, aiming at the imbalance of data distribution caused by CHT's inherent property, modify the distribution rules of virtual node on the hash ring, change mapping relationship between virtual nodes and real nodes, so as to change the distribution mode of data in a storage server, increase the balance of data distribution. Then, according to the characteristics of distributed storage system and data read and write access of storage system, design system runtime access rules and formula of load measurement, count and calculate each physical storage node load over a period of time, which as a foundation of judgment for subsequent load migration. Combined with the characteristics of multiple copies in a distributed block storage system, depending on the load level of the node storing copies, select a copy which read request. At last, on the basis of load difference of each node, combining with the characteristics of adopting virtual node when the data distributed, design three steps to carry out data storage including virtual nodes selection, migration and exchange, complete the migration of node load from high to low, making the system access load balanced.Based on open source distributed block storage system sheepdog to implement, eventually the two data sets of test show that new technology increases by 53.2% and 64.4% in the system access balabce, and makes system response time is reduced by 25.9% and 17.6%, compared with the original system without the use of the new scheme. The overall load balancing effect is obvious.