Bandwidth Load Balancing Algorithm In RAMCloud

With the advent of the Internet age, to social networks, represented by the rapid development of online data-intensive applications, on-line data for data-intensive real-time, interactive, accuracy, and disk access existing computer system uses a high delay, has become the era of big data bottleneck impact computer performance. MapReduce, NoSQL databases and other new systems and the emergence of cloud computing, with its virtualization, distributed processing, greatly improving the service performance of the computer system, but have failed to solve the problem of computer disk latency fundamentally.As memory prices continue to decrease in recent years, characterized by the rapid transmission of becoming a selection of alternative disk. Cloud RAMCloud memory is as a cloud system server storage media, it will cloud system data center migration from memory from disk into memory, fast data access and system memory, disk access delays abandoned big disadvantage, from a fundamental solution to the crisis disk bottleneck. But the memory of this special cloud cloud server, due to frequent access to data on the server memory bandwidth at a relatively high state. Data access frequency differences exist, there will be no memory cloud server load balancing memory bandwidth, high memory bandwidth at the server load, latency and high energy high bandwidth server resource utilization in low memory bandwidth load is not high, resulting in a server cluster latency and increased bandwidth and power consumption compared with the use of resources is not balanced.RAMCloud data centers stage log-structured file system, access frequency data segment determines the memory bandwidth of the server load, presents a DSE(Data-Segments Exchanging) algorithm based on the feature article, periodically the memory bandwidth of the data segment of the server load is not balanced, high access frequency selected data segments on high memory bandwidth load on the server and on the low memory bandwidth load on the server less frequently accessed data segment for mutual exchange, to reduce the high memory bandwidth load server bandwidth load, increased bandwidth load low memory bandwidth load on the server until it reaches the bandwidth load balancing cluster. Simulation results show that, DSE algorithm can load the memory bandwidth cloud cluster equilibrium, delay reduced 12.61%, and the DSE algorithm experimentally confirmed RAMCloud cluster energy consumption reduced 34.28%, effectively improving the memory of cloud clusters performance.