Method for dispatching I/O of asymmetry virtual machine based on multi-core dynamic partitioning

Abstract

The invention provides a method for dispatching I / O of an asymmetry virtual machine based on multi-core dynamic partitioning, which comprises the following steps: when a system is started, firstly an I / O dispatching initialization module completes the configuration of dispatching parameters and the allocation of a buffer zone; subsequently, when the system runs, an I / O status monitoring and information processing module dynamically collects and processes the runtime parameters of driving Domain and customer Domain, carries out statistics on parameters such as types, time and frequency of happened I / O event, and then sends the statistics result to a dispatching decision module for judgment; and the dispatching decision module makes dispatching decision according to preset conditions, and gives out commands to a multi-core dynamic partitioning module to execute the operation of multi-core dynamic partitioning or recovering, thus changing the core partitioning mode and dispatching statusof a processor of the system and leading the using mode of processor resources and the dispatching switching strategy of Domain to better adapt to I / O load requirements of the current system, and achieving the purpose of optimizing I / O performance.

Description

technical field [0001] The invention relates to a virtual machine scheduling method, in particular to a multi-core platform-oriented, asymmetric virtual machine I / O scheduling method, belonging to the technical field of computer operating systems and virtualization. Background technique [0002] The continuous development of computer hardware technology makes the performance of the server increase rapidly. In order to make full use of system resources under the premise of ensuring security and isolation, virtualization technology emerges as the times require. The basic idea is to use a thin layer of virtual machine monitor (VMM) software to The underlying hardware resources are managed, and at the same time, multiple isolated virtual hardware execution environments (VMs) are provided to the upper layer, so that multiple different operating systems can run on the same physical host. As early as the 1960s, IBM first used system virtualization technology in the 360 ​​series mai...

AI Technical Summary

Problems solved by technology

But at this time, the I / O request response delay caused by scheduling will cause a serious decline in the performance of I / O-intensive applications.

Although using self-virtualized devices and VM-bypass device access methods can effectively improve I / O performance, for most common devices that need to be shared, this method cannot solve the performance bottleneck caused by scheduling delays

Some existing virtual machine scheduling methods provide relevant optimization strategies, such as the Boost / tickle mechanism and the Partial Boost mechanism of the Xen system. This type of method can better meet the I / O performance bottleneck problem under the condition of a single application type, but In the face of complex application types, such as I / O computing mixed loads, it is still difficult to obtain ideal I / O performance with this method; moreover, most of the existing virtual machine systems adopt a globally unified scheduling strategy, and each processor core The scheduling decision-making conditions are consistent, but due to the diversification of the actual operating load types, the resource demand characteristics are different. At this time, it is difficult for VMM to provide an effective means of resource allocation.

In addition, most of the existing virtual machine scheduling algorithms are designed for single-core systems, seldom considering the application scenarios of multi-core environments, and cannot make full use of the characteristics of multi-core structures. However, multi-core and many-core have become the development trend of future hardware platforms. In this environment, the effective use of system processor resources to achieve the purpose of optimizing system performance is also one of the key issues to be solved in the design of virtual machine scheduling algorithms

Images