Study On Load Balancing And Performance Optimization Based On Redis Distributed System

With the rapid development of Internet,the data scale becomes bigger.Relational database does not dealing with large-scale data oriented service well,the NoSQL database represented by Redis is praised by the industry for its easy extension and high performance.There are two problems,the realtime load imbalance and response time,affecting the performance of NoSQL.The first problem make a single server node load larger and become bottleneck of the system.The second problem can lead to a loss of performance.Hot data is the main cause of real-time load imbalance.At present the main solution to solve the problem is the copy.However this solution need additional memory to store copies and CPU resources to maintain the consistency of the data.This paper proposes an adjusting load mechanism automatically without a central coordinator,which monitors cluster load though flow method and redistribute the hot data to achieve dynamic load balancing.Moreover,this paper proposes a sub-group balance strategy,which determines whether to load regulation and how to realize the load balance and data migration.In the design of decentralized distributed systems,each response of a request requires two TCP connections and multiple link shake will happen between requests,which can make system service access delay too long.To solve that problem,this paper proposes an adopt the cached map-group request strategy.This mechanism consists of mapping and submit requests batch.According to the mapping relationship,the client can directly find the correct service node,reduce the connection at a time.According to the target node,group some request and submit them to reduce shake between requests.