Data Acess Load Balance Of Parallel File System

With more and more high performance computing are needed ,parallel transaction is regardeg as a important method in high performance computing,also become important.In many cases,parallel processing transaction ability is weaken because of load imbalance,there are bottleneck in some action,and lower it's thoughput.So load balancing is important for maintain system performance.In order to carry out data accessing load balancing in parallel transaction,data migration and duplicate are used.Some model are added in PVFS,the load information are collected in every IOD,MGR sends request to IODs periodicity for load information.If detecting the heat of some disk high,finding the heat pot on the disk. Considering migrating those heat pot, The benefit/cost must be considered in migrition ,migrating the heat pot to the coldest disk,balancing the load.Migrating is migrating the heat pot from the hottest disk to the coldest disk. The system's module include:load balancing meta data management,load information's static,collection,load balancing computing,load migrition. After load balancing in PVFS,the performance has some rising.Some data's heat is so high that the disk's heat become very high when the data is migrated in,a new load imbalance appear.Duplicate can solve this problem,when the heat file is accessed,the heat file already has several duplicate,randomly selecting a duplicate to read.Duplicate can solve all load imbalance,but the duplicate's consistency must be kept,the duplicate must be read in parallel and write in chain.