Support High Performance Ipc Memory Management Strategy Research

With the development of operating system technology, multi-task has become one of essential features. More than one tasks running on a same computer provides users convenience and flexibility and allows many jobs to share resources on one single computer. In the meantime, processes need communicate with other ones for completing related work, which need features like inter-process communication (IPC) provided by operating systems. Better IPC performance means shorter responding time, which could have tasks complete more efficient. Starting from memory management strategy and implementation method, costs during IPC are analyzed in this thesis to find out ways to increase IPC performance on Linux based on modifying memory management strategy.First, it is introduced in this thesis that background and worth of this study, as well as related domestic and oversea researches. Requirement of memory manager in modern operating systems and inter-process communication problems on such architectures are demonstrated. Several types of OS kernels and appropriate memory management strategies are described. Different IPC mechanisms for different kernel design models are also introduced.Next, several IPC mechanisms and implementations on representative OS kernels are drawn to attention. Memory structure and IPC methods for user land processes of Linux kernel, a typical monolithic kernel, are carefully studied. It is analyzed how Linux IPC is implemented by reading related source code. Differences of IPC methods and their cost are also summarized. It is then analyzed that IPC methods for system services of MINIX3and L4, both of which are micro kernels. After reading research materials of these kernels, analyzing their design principles, and reading related source code, their IPC mechanisms on different memory models are studied.After comparison of previous IPC methods and analysis on ways of improving Linux IPC performance, it is tried that bring L4’s memory management strategies to Linux and applying IPC with direct memory mapping cross process address spaces. Sematic of several operations for direct memory mapping is also designed. Then, feasibility of such mapping on Linux kernel is analyzed. Detailed implementation on Linux kernel, as well as its impact to other part of the kernel, is also introduced.Finally, several tests of the direct memory mapping against existing Linux IPC methods on their transmission speed and single transmission overhead are done. Results show that direct memory mapping has shorter overhead than existing memory sharing method does for a single transmission, while it can transmit as fast as the latter does for large amount of data.