Whitepaper: Performance Isolation of a Misbehaving Virtual Machine with Xen, VMware and Solaris Containers
Thursday, February 23, 2006
| 2 Comments
|
A group of researchers at Clarkson University produced an absolutely interesting paper about a rarely analyzed aspect in virtualization technologies comparison: virtual machines isolation:
...how well do different virtualization systems protect VMs from misbehavior or resource hogging on other VMs? In this paper, we present the results of running a variety of different misbehaving applications under three different virtualization environments VMware, Xen, and Solaris containers. These are each examples of a larger class of virtualization techniques namely full virtualization, paravirtualization and generic operating systems with additional isolation layers. To test the isolation properties of these systems, we run six different stress tests - a fork bomb, a test that consumes a large amount of memory, a CPU intensive test, a test that runs 10 threads of IOzone and two tests that send and receive a large amount of network I/O. Overall, we find that VMware protects the well-behaved virtual machines under all stress tests, but sometimes shows a greater performance degradation for the misbehaving VM. Xen protects the well-behaved virtual machines for all stress tests except the disk I/O intensive one. For Solaris containers, the well-behaved VMs suffer the same fate as the misbehaving one for all tests.Read it here. It's a real real real pity this wonderful comparison didn't included Microsoft Virtual Server. This could be a critical point in products comparison, apart any marketing talk. I hope to see a second version soon.
© 2003-2008 virtualization.info
Sitemap
Sitemap


Comments
While this is an interesting summary of conclusions, without the actual configurations of their tests it is pretty meaningless. For example, there is no mention of any solaris container resource management controls being applied to there zones. If this is the case, it demonstrates a complete lack of understand about Solaris containers operate. A similar problem with VMware -- why did they use Worstation ? People hosting apps use ESX or the newer VMware server.
By
/k, at Thursday, February 23, 2006 6:46:00 PM
Thanks all for your interest in our paper.
The paper pointed to is actually a submission to USENIX. We could include more information on our configurations in the actual paper. That is good feedback.
We didn't use VMware ESX or other servers because my understanding is that their EULA prevents publishing such results.
The paper contains some notes about the resource managment controls we tried with the Solaris containers. If you have some suggestions of controls to try that we did not, we would love to hear them.
Jeanna Matthews
Clarkson University
By
Jeanna Matthews, at Friday, February 24, 2006 12:44:00 AM
Post a new comment