Structured Peer-to-peer Network Load Balancing Algorithm Based On Access To The Hot Spots

Structured peer-to-peer network (P2P) is one of the new technologies in recent years, its appearance has changed the traditional server-centric service model, and it has make people share resource more convenient and more efficient. However, structured peer-to-peer network is designed and implemented at the assumption that the popularity of all files is the same. However, in the actual network, the distributions of queries are very uneven, which makes a lot of nodes visit some nodes frequently, and then facing emergency visits, it is easy to cause query Hotspot problem. The emergence of query hotspot will decline the performance of the node which can't provide effective services, or even affect the entire peer-to-peer networks's quality of service.In peer-to-peer network, node is mutual equal and there has no centers exist, the traditional center-based server load balancing scheduling algorithm does not apply for it, so studing new load balancing algorithm based on the characteristics of the the need-to-peer network is necessary.In this paper, the author has analysised and researched the peer-to-peer load balancing technology both at home and abroad, and proposed three load balancing algorithms to solve query hotspots. Replication technology is often used to deal with query hot spot, but when the Replication technology select reproduction node, it depends on certain fixed nodes more excessively, and it also do not take over the load and utilization. In view of these deficiencies, this paper presents an improved load balancing algorithm which is based on the replication.Through experiments,it is proved that the improved algorithm in the load transfer cost and the utilization of node have improved.After studing the characteristics of load, we draws that node's load is predictabe, and Combination of time-series approach we proposed load balancing algorithm based on time series .The algorithm based on node's past load value predict the node's future load in a period of time, and the time series algorithm will transfer the hot files to reproduction node before the node's load is overweight. When the algorithm select the reproduction node,it will consider the remaining load,the increasing load and the average online time,which will avoid the load by frequently transfering reproduction node .Through experiments, t is prove that the time series load-balancing algorithm can effectively predict node's load, and can reduce the average load transferring cost.After researching on the time of transferring the node's load, this paper presents an active load-balancing algorithm which is based on the query rate. Therefore The algorithm according to the query rate actively hash the hot files to reproduction node.Taking into account there are many queries in a short period of time when the hot spots occurs, so algorithm Uses Muti-windows Services model. At the same time, the query rate threshold is dynamically setted, which can dynamically set the hash speed of the hot resource in accordance with the node's load.Through experiments,it is proved that the algorithm can effectively reduce the node's load and the average respond time of the query.