Design And Implementation Of Resource Management For Multikernel Operating System

With the rise and development of cloud computing technology and big data,there is a very serious challenge with the reliability,scalability,implementation efficiency and the performance of the data center operating system.The traditional data centers use virtual machines for resource management.Virtualization technology transforms a large number of traditional physical resources into logical and manageable virtual resources.For example the single processor will be virtualized as a multiprocessor.The virtual machines have good performance isolation and security isolation.As a result,the virtualization technology is an effective way of resource management.With the increasing of the core numbers in the current data center hardware platforms,due to the design of the lock,the shared data structure,the shared memory and the other traditional mechanisms,there will be a scalability issue on the platform,which will bring the serious performance overhead.Therefore,the issue how to improve resource utilization and reduce operating and maintenance costs become the focus of the current virtualization technology research.Rainforest,a typical multi-kernel operating system designed and completed in this thesis,is a method of resource management on a multi-kernel operating system,which will do the operation of the reasonable and effective physical resource allocation and the corresponding resource management in a physical node.With this way it can implement the isolation and management of the physical resources(CPU,memory)on the many-core platform.Finally we can create and destroy a sub operating system on this resource partition.In the multi-kernel operating system resource management method,the physical resources are allocated directly in the management resource pool,and the partitioned sub operating system is deployed on the partitioned hardware resources.All the sub operating systems have the exclusive physical resources and address space which won't be shared with each other.At the same time with the comparison with the virtualization proposal,due to this method is without hypervisor layer overhead,so it has the following characteristics,strong isolation,high flexibility and small additional performance overhead.This method aims to designed to run large data mixed load to improve resource utilization when we run the big data mixed load on the many-core platform.