| In recent years, due to the concept of Internet of things, the recommendation system based on big data and so on, a new challenge is put forward to the storage of high concurrent and massive small data.The more commonly used distributed system designed with a central node is abandoned in this paper and a P2 P structure based on Gossip protocol is chosen. In the central node structure design, there is a single-point failure and data hotspot issue which has a great requirement to the cost of system hardware and maintenance. From the view of request distribution and cost reduction, around the P2 P structure, we study many mature technology from current industry like node synchronization in P2 P system, node failure detection, data distribution strategy, redundant data copy policy and so on. To improve the ability of concurrent processing, we discuss how to efficiently complete the message transmission, message distribution and the management of data storage in the single node.Here is the brief introduction of this paper below:(1) Gossip protocol in the communication between P2 P nodes is researched and realized in this paper.(2) Learning by Phi Accrual Failure Detector algorithm, a module to detect failure node is designed by establishing sample space and using probability distribution modeling theory which can be more effective to reduce the false detection.(3) A simple data distribution policy is discussed and designed. The process of client read/write request is simplified to reduce the time delay.(4) The Reactor Model to process event in the system is used in order to improve the ability of concurrent processing. Multi-thread techniques are also talked about to give full play to the advantage of multi-core or multi-processor.(5) The optimization techniques of storing data in the single node is studied and implemented like transversely cut file, memory mapping, make full use of sequential read/write in the disk to improve the efficiency of the I/O. |
PDF Full Text Request