The Design And Implementation Of An Efficient Memory Deduplication System Based On UKSM

Reducing data redundancy is one of the central problem in server.All kinds of service need more and more memory.Traditionally cloud computing creates Virtual machine(VM)to support customs' need.Different virtual machines contains the same service or files which lead to memory redundancy.Besides,technology of containers becomes mature,processes and files between different containers and host also contain redundant data.Personal computers and mobile phones also have this dilemma.Ap-plications including office,images handling,3D presentation and so on need a large amount of memory.It is promising to develop a better way of reducing memory redundancy.computer system may have a lot of redundant memory which are completely the same or partially the same.current research has proposed memory deduplication and memory compres-sion,which are already implemented in current Linux kernel.Memory deduplication is to remove the identical memory,which allows memory keeps only one copy.Memory compression compresses memory to a much small size.Unfortunately,these two meth-ods are designed by two different groups,and they are based on different princeples,In this condition,they can not be replaced,in order to make them work concurrently,we need a new mechanism.Reducing redundancy of file cache is always a difficult problem.File system keeps file cache of each file in memory.So Reducing redundancy of file cache helps to save memory.This paper has following contributions:1.We implement our design to make the two methods work and coordinate with each other,we takes system delay,response time and CPU expand into consid-eration.We have realized our design in Linux kernel.2.We have tested this new mechanism with a set of real applicaations obtaining a significant performance improvment.3.We develop a method to combine identical page in file cache.This method takes use of the UKSM scanning method.Our contributions include Splitting file cache to smaller pieces,designing a interrupt-safe way for comparing and merge pages and designing a copy-on-write method for merged file page.4.Design experiments to test Memory deduplication method.In our experiments,the average rate of merging file pages is up to 2.28MB/s.