Xen VS OpenVZ Round Two? HP's Performance Evaluation of Virtualization Technologies for Server Consolidation By Trevor Connors published: Tuesday, April 24 2007
Updated: XenSource adds comments - 4/30
Xen VS OpenVZ Round Two? OpenVZ project manager Kir Kolyshkin and Simon Crosby of XenSource both add their comments.
I believe Simon Crosby's response to a paper with similar content and conclusions should again be revisited.
OpenVZ project manager Kir Kolyshkin seems to agree as well, he wrote:
"Results from the HP labs clearly show that OS level virtualization technologies like OpenVZ are the definitive way to consolidate servers, due to much lower overhead, higher performance and better scalability than hypervisor-based technologies.
Of course, hypervisors serve an important role and are required if someone wants to run different OSs on the same piece of hardware. So, both technologies have their applications, and in fact they can both co-exist."
Simon responds:
"It has always been known that in the case where all applications run against exactly the same kernel version, that a container type technology such as jails/ OpenVZ/ Virtuozzo/ Solaris Containers (a) saves memory through code re-use of the core OS (b) is likely to be more efficient in cache use (same code base, after all) and (c) will perform very well."
"This is the (only) sweet spot for container technology. But try asking Virtuozzo for their performance simultaneously running RHEL 3, Windows Server 2003 and Vista, and you'll be met with a blank stare - the technology cannot host a mixed workload of different OS types and so isn't useful for typical server consolidation, HA, DR or other virtualization enabled use cases. That's why Sun is adopting Xen for Solaris 10 on x86 - to serve the key enterprise use cases for virtualization."
"The HP comparison is another badly formed comparison of apples and oranges."
Even though I still prefer pineapples I would like to thank Simon and Kir for helping us understand, not only the comparisons but more importantly, the differences in these two technologies.
VSM will continue this topic in an upcoming podcast with SWsoft about "OS virtualization and how it is different from hypervisor based virtualization"
Below is a sample from a PDF named "Performance Evaluation of Virtualization Technologies for Server Consolidation" found here on HP's website.
Introduction
There has been a rapid growth in servers within data centers driven by growth of enterprises since the late nineties. The servers are commonly used for
running business-critical applications such as enterprise resource planning, database, customer relationship management, and e-commerce applications.
Because these servers and applications involve high labor cost in maintenance, upgrades, and operation, there is a significant interest in throughput and
response time, impacted compared to its performance on a base Linux system?
-
As workload increases, how does application-level performance scale up and what is the impact on server resource consumption?
-
How is application-level performance affected when multiple tiers of each application are placed on virtualized servers in different ways?
-
As the number of multi-tiered applications increases, how do application-level performance and resource consumption scale?
-
In each scenario, what are the values of some critical underlying system metrics and what do they tell us about plausible causes of the observed virtualization overhead?
The remainder of the paper is organized as follows. Section 2 discusses related work. Section 3 introduces the architecture of our testbed and the various tools used. The details of the experiments are described in Section 4, and the results along with our analysis are presented in Section 5 and 6. Finally, we summarize our key findings and discuss future work in Section 7
....
14 pages not shown
.....
Conclusion
In conclusion, there are many complex issues involved in consolidating servers running enterprise applications using virtual containers. In this paper, we evaluated different ways of consolidating multi-tiered systems using Xen and OpenVZ as virtualization technologies and provided quantitative analysis to understand the differences in performance overheads. More work can be done in extending this evaluation for various other complex enterprise applications, including applications with higher memory requirements or database-intensive applications. We hope that systems researchers can use these findings to develop optimizations in virtualization technologies in the future to make them more suited for server consolidation.
download the full PDF Here.
Read the latest on this topic in this VSM articleXen VS OpenVZ Round Two Continued - Updated 7-2-07
Read what Simon Crosby (CTO of XenSource) responded to Xen vs OpenVZ comparisons in the past
Listen to the Podcast: Interview with Simon Crosby, CTO and Founder of XenSource
More on Xen Here
More on SWsoft Here.
|
