Data Migration Strategy For RAMCloud Hierarchical Hybrid Storage Architecture

With the advent of the information age, online applications with intensive data represented by social networks have been developing rapidly. Intensive online data have higher requirements for instantaneity, interactivity and accuracy. However, there is high latency in the existing storage system with hard disk access, which has become the bottleneck of computer performance in the era of big data. The service performance of the computer system has been improved greatly by the emergence of cloud computing, NoSQL databases, virtualization and distributed processing while the problem of computer disk latency has not been solved fundamentally.In recent years, with the development of hardware technology, memory with characteristics of fast rate of transmission is becoming one of the choices to replace disk. RAMCloud is a cloud system that uses memory as its server storage medium, which moves the data center of the cloud system from disk to memory, achieves fast access to data in memory system, casts the disadvantages of great disk access delay away, and solves the bottleneck crisis of disk access fundamentally. The high price and lower capacity ratio of RAM make the cost of the RamCloud cluster too high, which leads to RamCloud application failure in a large scale to the data storage of mass online intensive application.In order to achieve the goal of high efficient storage, access to the huge amounts of online data, and increase the data storage capacity of RamCloud, the idea of hierarchical storage management is proposed in this paper. By introducing SSD as the second level storage medium, an adaptive migration strategy based on the importance ranking of data is proposed which constitute three stage of hybrid storage architecture with Ram and HDD. Firstly, the importance of data itself is calculated based on factors like the size of data itself, the importance of time, and the total times of user access and so on. Secondly, the potential value of the data is evaluated by adopting the method of users’ similarity in recommendation system and the importance ranking of PageRank algorithm. The importance of the data was determined by the importance of data itself and its potential value. Then, an adaptive data migration strategy is designed based on the importance ranking of data. The experimental results show that, the proposed model can identify the importance of data, place data in a hierarchical way and improve the data access hit rate of the storage system compared with algorithms of Least Recently Used(LRU), Least Frequently Used(LFU) and Migration Strategy based on Data Value(MSDV). Meanwhile, the proposed model can alleviate the storage pressure of massive data in memory cloud system and achieve the goal of reasonable migration placement of hot data.