Research On Some Key Algorithms In A Distributed File System

With the rapid development of the social, economic, Internet, global annual data growthes at geometric orders, we have entered the era of big data. How efficiently and reliably storage and access to these data is the business community, academia, research hotspot. Past data storage are centralized storage which with the rapid growth of the amount of data and the increasing storage demand for access computing resources,IO resources, network resources and storage resources will be a bottleneck.Over the past decades, people realized that using a large number of cheap computers composed of a cluster can solve the growing storage, computing, network access requirements. Now, the cluster has become the mainstream high-performance computer architecture. Build a distributed file system in the cluster is the mainstream solution to solve massive data storage and access.This article has been studied for a number of key issues in the distributed file system for cluster computing, the main work is as follows:(?) Proposed and implemented a lock-free algorithm for network message queue. Currently the most popular distributed file system server is multi-threaded. The network thread communicates with other threads through a lock-based message queue. With the increasing access, the lock conflicts are a performance bottleneck of the network module. This paper presents a lock-free network message queue algorithm to solve the performance problems in one-reader-one-writer environment.(?) Proposed and implemented a pipeline-supported data server access algorithm. Data in the distributed file system is decentralized stored in different data server. Most of the current distributed file systems use the lock manager for multiple-data access. This paper presents a new distributed mutual exclusion algorithms take full advantage of the distributed cluster of data access in the "pipeline", so to maximize resource utilization, effective solution to the the lock management server access latency issues.(?) Proposed and implemented a consensus algorithm in dynamic systems. Consensus algorithm is one of the basic algorithms used in the distributed systems. The traditional distributed consensus algorithms, such as Paxos algorithm, do not to support dynamic systems. This paper presents an improved Paxos algorithm, which can support dynamic systems.