Memory Analysis Towards More Efficient Live Migration of Apache Web Server

Open Access
Cao, Wenqi
Graduate Program:
Computer Science and Engineering
Master of Science
Document Type:
Master Thesis
Date of Defense:
March 27, 2015
Committee Members:
  • Peng Liu, Thesis Advisor
  • Guohong Cao, Thesis Advisor
  • Virtual Machine
  • Live Migration
Virtual Machine live migration is a key technique in the clouds. It can benefit for the workload balance and data backup. During the virtual machine live migration, memory is an important part that needs to be transferred from the source to the destination. Current techniques can be divided into two categories: Pre-Copy and Post-Copy. Both approaches migrate all memory blocks from source to target without memory analysis before live migration. However, some memory blocks are not necessarily transferred through live migration process, because the content of these memory blocks always remains the same. Moreover, modern operating systems use unused memory to cache recently accessed blocks of permanent storage device. This technique is called PreFetcher or SuperFetch. Although Prefetcher can speed up the amount of time that system takes to start up programs, it occupies more memory spaces which might not be used by users. These observations lead us to propose an approach to analyze the Virtual Machine memory page table and divide the memory data into two parts (active memory and relatively stable memory). Instead of migrating all Virtual Machine memory through live migration, we only live migrate the active memory. The relatively stable memory can be merged into Virtual Machine image file. Thus, during startup process, target machine can directly load these memory blocks into its physical memory through image file stored in shared storage. The proposed approach can considerably reduce the amount of data sent between the two hosts involved in the live migration and has the potential to shorten the total migration time.