Introduction to Space-Efficient Checkpointing Xen

Overview

Our software is fast and space-efficient.
With our software, you can use faster and more efficient checkpoiniting technology.

A checkpoint, in a virtualization context, is a snapshot of the state of virtual machine.
Typically, a full image of the VM's memory along with processor and device states are recorded.

In contrast to existing methods, our technique eliminates redundant data and stores only a small portion of the VM's memory. We accomplish by transparently tracking I/O operations of the guest to external storage and maintaining a list of duplicated memory pages. As a result, these pages are excluded from the checkpoint image at a checkpoint.

 

<Figure 1>  Disk I/O flow in the Xen VMM and the page-to-block map

 

Results

 We have conducted experiments on HVM environment by using 'make' and 'Postmark' benchmark.
We used optimized Xen 4.5 and guest Linux as version 3.13.11-ctk21+ on the SSD.
In the first, a single checkpoint was taken after each benchmark application was run to completion.
And in the next two rounds, a series of checkpoints were taken at fixed interval as a state which it just like when I ran benchmarks.

The graphs below are the results of our experiments.

  Comparing our method to the unmodified approach, we achieve a 60.03 % reduction for the Make and 60.82% reduction for the Postmark in the time consumed to save the checkpoint.

 

Publications

  • Eunbyung Park, Bernhard Egger, and Jaejin Lee. "Fast and space efficient virtual machine checkpointing". In Proceedings of the ACM SIGPLAN/SIGOPS International Conference on Virtual Execution Environments (VEE'11), Newport Beach, USA, March 2011.
    [pdf] [bibtex] [doi]
  • Bernhard Egger, Erik Gustafsson, Changyeon Jo, and Jeongseok Son. "Efficiently restoring virtual machines". Presented at the IFIP International Conference on Network and Parallel Computing (NPC 2013), Guiyang, China, September 2013, in Springer International Journal of Parallel Programming (IJPP), Volume 43, Issue 3 (June 2015).
    [pdf] [bibtex] [doi]

 

Download

 

Contributors

  • Eunbyung Park, Changyeon Jo, Younghyun Cho, Jeongseok Son, Hochan Lee, Erik Gustafsson, Bernhard Egger

 

Contact and Bug Report