Dynamic Huge/small-page Memory Adjustment Based On PHPA

In order to improve the utilization of physical resources,virtualization techonology has been widely used.With virtualization technology,a single server can integrate many virtual machines to provide users with a variety of services,in this way,in order to ensure a good user experience,this server will be in great need of hardware resources,including memory recources.Currently,we can use huge page framework which is based on hugetlbfs to provide 2M hugepage for service applications,thus ensuring service performance,but it does not support memory compression,and its scalability is very poor Our lab's previous work has achieved a hugepage compression system based on PHPA(Pristine Huge Page Allocator)called Hzmem(H stands for huge,z on behalf of the compression)which compresses the hugepage memory and stores them back to memory,thus increases the available hugepage memory in some way.Inspired by the design of transparent huge pages,in some real scenarios,for example,the server's needs for huge pages is far more than the demand of small pages,but there are only free small pages left and no huge pages.On the basis of the existing achievements of our laboratory,this paper implements a method of dynamic adjustment to memory pattern based on PHPA huge page framework to solve the above problems.The main contributions of this paper are the followings:1)Based on the PHPA huge page framework,a significant reduction of the metadata overhead has been made,saving a lot of physical memory resources.2)Achieved a way of dynamic adjusting the size between huge page and small page,very flexible.3)There is no need to modify the original Linux kernel,we implement our method as a module,which can be used after it is inserted to the running kernel.4)We provide a user mode interface to modify the memory pattern,which is very convenient.Through the functional and performance tests,this paper realizes what we expected and loses little performance of the original system.