A Task Guided Framework for Reducing the Service Downtime of Virtual Machine Live Migration

• Main Authors: KUO, YUAN-HAN, 郭遠翰
• Other Authors: LIN, CHO-CHIN
• Format: Others
• Language: zh-TW
• Published: 2018
• Online Access: http://ndltd.ncl.edu.tw/handle/7cvv6u

碩士 === 國立宜蘭大學 === 電子工程學系碩士班 === 106 === Virtual machine live migration is an important mechanism for cloud computing to achieve non-stop services. Different from normal migration, the live migration with pre-copy scheme employs two stages of page synchronization. In the first stage, memory pages are selected according to the pre-copy rules and those selected pages are synchronized from the source host to the target host. The pre-copy rules are designed to reduce the number of synchronized pages in the second stage: the stop-and-copy stage. By reducing the number of synchronized pages, the service downtime can be minimized and the quality of service can be ensured. The service downtime of live migration is affected by page-accessing frequency and the size of working set. Therefore, the users may experience a long service downtime if a migrating virtual machine enters the stop-and-copy stage at an inappropriate time point. This thesis proposes a task-guided framework that is useful for reducing the service downtime when a live migration is conducted. The task-guided framework allows a program developer to tag task segments that are suitable for migrating a virtual machine with a short service downtime. However, a task segment may have been completed before a virtual machine enters the pre-copy stage. In this case, a large number of pages may need to be synchronized across hosts. In order to avoid the situation, an interrupt mechanism is added to prevent a virtual machine from entering the stop-and-copy at an inappropriate time point for an ongoing live migration. The interrupt mechanism is achieved by asking the hypervisor to suspend the live migration when the program exits a tag segment. The pending live migration will be resumed when the program enters the next tag segment in order to ensure a short service downtime.