Workload-Aware Live Storage Migration for Clouds

Overview

The emerging open cloud computing model will provide users with great freedom to dynamically migrate virtualized computing services to, from, and between clouds over the wide-area. While this freedom leads to many potential benefits, the running services must be minimally disrupted by the migration. Unfortunately, current solutions for wide-area migration incur too much disruption as they will significantly slow down storage I/O operations during migration. The resulting increase in service latency could be very costly to a business.
This research presents a novel storage migration scheduling algorithm that can greatly improve storage I/O performance during wide-area migration. Our algorithm is unique in that it considers individual virtual machine's storage I/O workload such as temporal locality, spatial locality and popularity characteristics to compute an efficient data transfer schedule. Using a fully implemented system on KVM and a trace-driven framework, we show that our algorithm provides large performance benefits across a wide range of popular virtual machine workloads.

People     

• Faculty: T. S. Eugene Ng

• Collaborator: Kunwadee Sripanidkulchai
• Students: Jie Zheng

Source code

• We implement the scheduling algorithm based on QEMU-KVM 0.12.50. 
• Source code of Qemu-scheduling-version-1.0 is released. (Download)
• Compilation
• ./configure --prefix=/your_path_for_placing_new_generated_executable_file,  e.g. ./configure --prefix=/usr/local/kvm
• make 
• make install
• More details about setting up VM and configure KVM are on KVM How-To.
• Command
• Select the qemu console (CTRL+ALT+2)
• Type "help migrate" for migration command help information. 
• migrate [-d] [-b] [-i] [-s] [-t] uri  -- migrate to URI
• -d, -b -i are the existing options in qemu-kvm 0.12.50
• -s and -t are the new options that we added
• -s: enable scheduling algorithm in storage migration
• -t: enable throttling mechanism in storage migration to make sure migration converge
• More details about migration options are on KVM Migration

Workload I/O Trace

• To investigate storage migration scheduling, we collect and study a modest set of VMware virtualization benchmark(VMmark) I/O traces. VMmark includes five types of servers which are representative of the applications run by VMware users. Five servers are mail server, file server, web server, java server and database server. The details about how to collect the traces are described in the documents below. For each workload, the READ and WRITE operations are logged into separated files. Each file includes three columns which represent access time, access offset and access data length. Note: In the documents below, we mentioned that "Each trace contains the I/O operations executed by a server over a 12 hour period".  The statement should be corrected as "Each trace contains the I/O operations executed by a server ranging from 5.6 hour period to 12.9 hour period".
• The trace files are released. (Download). 

Documents

• Workload-Aware Live Storage Migration for Clouds, Jie Zheng, T. S. Eugene Ng, Kunwadee Sripanidkulchai; ACM Virtual Execution Environments '11, Newport Beach, CA, March 2011

Funding

This research was sponsored by NSF CAREER Award CNS-0448546, NeTS FIND CNS-0721990, NeTS CNS-1018807, by an IBM Faculty Award, an Alfred P. Sloan Research Fellowship, and by Microsoft Corp.  Jie Zheng is additionally supported by an IBM Scholarship. Views and conclusions contained in this research are those of the authors and should not be interpreted as representing the official policies, either expressed or implied, of NSF, IBM Corp., Microsoft Corp., the Alfred P. Sloan Foundation, or the U.S. government.

Contact