Dissertation > Excellent graduate degree dissertation topics show

Lock-based multiprocessor -aware scheduling system VCPU

Author: GengZuo
Tutor: WuSong
School: Huazhong University of Science and Technology
Course: Computer System Architecture
Keywords: Multiprocessors Virtualization Spinlocks Lock holder preemption Lock perception VCPU Scheduling
CLC: TP332
Type: Master's thesis
Year: 2011
Downloads: 8
Quote: 0
Read: Download Dissertation

Abstract

Multiprocessor system, the introduction of virtualization, will bring great advantages - multiprocessor system makes virtualization technology can be applied more may otherwise be wasted processor cycles and computing resources, enabling centralized computing and resources dynamic allocation, full use of the system resources, performance to a new level. But at the same time, the introduction of multi-processor systems virtualize, it also raised a number of issues, the lock holder preemption issue is one of them. From the design point of view, the spin lock can only be held for a short time. However, if a virtual machine VCPU be preempted when a lock is held, then it is no longer up - the other virtual machines running VCPU may apply and wait for the lock release, which led to the lock hold time extension, a large number of processor cycles are wasted, greatly affected the performance and scalability. VCPU multiprocessor lock aware scheduling system provides a virtual multiprocessor environment VCPU scheduling policy. System through the kernel in a virtual machine monitors spin locks, spin lock for accurate counting and counting results based on Hypervisor (virtualized environment management and monitoring systems) used in an efficient and fair scheduling policy, effective solution lock holder preemption problems and improve the multi-processor system virtual machine performance. System monitoring module locks the kernel from the virtual machine in the virtual machine to obtain the lock count information, and passes it to the communication module in the middle layer; communication module will get a summary of information, and data classification and mapping, by all the VCPU virtual machine locks on each count information; Hypervisor layer VCPU scheduling control module reads the lock count information to make VCPU control decisions; virtual machine statistics module from the Guest OS kernel to obtain the original lock status information data, and data into a readable classification lock status data statistics, print to the console, or generate spin lock statistics log files. Function test results show that the system to normal operation, during operation, the lock can be avoided when scheduling VCPU holder preemption. Performance analysis is mainly the original Xen systems and the use of systematic comparison of existing solutions IO intensive computational load and CPU load runtime performance. Performance test results show that, compared to existing solutions, the system is running IO intensive workloads performance increase of about 10%, operating performance CPU computational load increases by about 5%. Both load running of the process, the average spin lock hold time and wait times dropped significantly compared to the original Xen system for IO-intensive workloads, a decrease of 28% and 97%, for CPU computational load, a decrease of 74% and 99%.

Related Dissertations

CLC: > Industrial Technology > Automation technology,computer technology > Computing technology,computer technology > Electronic digital computer (not a continuous role in computer ) > Arithmetic unit and the controller (CPU)
© 2012 www.DissertationTopic.Net Mobile