Design And Implementation Of A Distributed Memory Management System Based On Page Fault

Virtual memory technology and request-for-page technology are used by the Linux operating system to manage physical memory.Only the pages needed for the present process are loaded and mapped to logical addresses in physical memory.Other pages'data that aren't being used right now is still stored on the disk.When memory is scarce,the Linux operating system employs a page swapping and scheduling technique to free up space in physical memory for more active pages.The Linux operating system usually moves anonymous pages to a swap region on disk,which has a much lower I/O rate than memory.As a result,if memory is limited,frequent page shifting can dramatically improve reaction time.Therefore,when memory is insufficient,frequent page swapping can significantly increase the response time of the system.As a result,the swapping of anonymous pages on disk-based devices becomes a performance bottleneck when there is not enough free memory left.To address these issues,this thesis investigates the technical principles and details of the kernel-based RDMA interface,remote memory management and block device drivers from the perspective of overall cluster resource utilisation,and writes an RDMA-based virtual block device and a remote memory service program.The RDMA-based virtual block device and a remote memory service program have been written to establish a mapping between the virtual address of the swapped anonymous page in the system's page table entry and the physical address of the remote memory,converting I/O requests for page swapping on Linux systems into read and write operations to remote memory using RDMA technology.A copy of the page data from the write operation is also saved to the local backup swap.A distributed memory management system based on page fault is thus implemented.The thesis then tests the availability and performance of the system,comparing the size difference in read and write throughput between the system and a normal swap area,and the difference in response time when running computationally intensive programs.The results show that the system guarantees good availability and high performance,and can effectively improve the memory utilisation of the cluster and significantly speed up the exchange rate of anonymous pages.