Research On Dynamic Extension Of Distributed Cache Systems

With the rapid development of internet technology, the number of network users has increased sharply and they have had higher demand for data interaction and faster service response. Application data that needs to be processed will be upgraded from PB level to EB level, and often had massive access to data, which has become a serious problem that needs to be dealt with. The data in distributed cache was managed by the form of Key-Value without using the original relational database. Distributed cache abandons the original relational database and manages the data by using the right form of the Key-Value button, while it is accompanied by high-speed memory as the storage medium. It can solve the I/O bottleneck problem between the background database and the application server effectively and improve the speed of data access.Firstly, this thesis analyzed the research background of distributed cache, compared some typical products of distributed cache and then introduced the characteristics, application scenarios and the process of development of distributed cache.Secondly, the research on dynamic scalability of distributed caching system is carried out, on the basis of consistent hash algorithm, it proposes a kind of dual mapping strategy based on virtual nodes to speed up the process of dynamic expansion. By mapping virtual nodes to both of the current and former server, this strategy ensures the hit ratio of the data request, at the same time, the system space is expanded and the performance is improved.Thirdly, regarding to the hot data in distributed system, rebalancing strategy was proposed to realize the load rebalancing. Rebalancing strategy can reduce the proportion of serious virtual node on ring and lengthen the light-load node by changing the position of virtual node on hash ring, and then relieve the problem of load unbalance.Lastly,the research build a distributed caching system by using Memcached to provide a caching service for an online bookstore. Test and analysis of the two strategies of this paper through simulating user requests by LoadRunner, the experimental results shows that VDM strategies having good dynamic scalability, it can finish extended tasks quickly. The displacement strategy of VLA can solve hot data issues effectively and balance the load of the system.