Query Hotspot Elimination Scheme Based On Replication In Structured Peer-to-Peer Networks

With the increasing capability of personal computer and popularity of network technology, Peer-to-Peer (P2P) computation mode, which aims to utilize every possible source in internet fully and reasonably, has been widely used in many fields. The implementation of structured P2P network assumes that all data items be of the same popularity. However, the distribution of queries for real data items has been shown to be highly skewed, with several popular objects being requested most of the time. This type of traffic may overwhelm the source nodes that host the frequently-accessed data items. When flash crowd happens, the amount of the requests for the popular objects can increase dramatically to tens or hundreds of times as compared with the original amount. Such nodes may suffer from severe performance failures, and almost all the services they provide will become unavailable. Replication technology is usually used to handle hotspot problem, which can reduce the number of dropped queries and ensure high network service quality.Replication technology includes: replica creation strategy and replica utilization and management strategy. By creating replicas for heavy-loaded nodes, we can transfer load to low-loaded nodes. Thus, load for heavy-loaded nodes can be reduced. Replication overhead is one of the most important problems in replication technology. This paper optimizes replication strategy, for the purpose of reducing dropped queries under the hotspot phenomenon and minimizing replication overhead.(1) Optimized replication time: By foreseeing traffic surges on a node. Replicas can be scattered before flash crowd happens, which consequently prevent nodes from overloading and dropping queries.(2) Optimized replication strategy: In order to distribute load fairly onto each node, we propose a fair replication algorithm. We choose light-loaded nodes along the query path to keep replicas. At the same time, we adaptively adjust load redistribution speed for heavy-loaded nodes, according to their load status. In this way, we can effectively control replication overhead, by keeping appropriate number of replicas. (3) Replica utilization strategy: this paper proposes a probabilistic replica selection algorithm, in which queries are forwarding to the low-load nodes with high probability. This complies with the principle that the one with more capacity should take more responsibility. In this way, nodes with lower load can be used to shed more load, therefore we can make full use of the replica to distribute load.We perform a series of experiments to validate our method. The results show that our method can effectively reduce dropping rate with lower replication overhead.