The DFS-N test team has completed some extensive Performance and Scalability testing and we wanted to share some of the results.
To start, here is a comparison of the performance of adding links to a standalone namespace. It includes data for Windows Server 2003, Windows Server 2008 and Windows Server 2008 R2:
As you can see, there is a significant improvement between Windows Server 2003 and Windows Server 2008. Windows Server 2003 goes “off the chart” (above 1 second) after around 30,000 links. Windows Server 2008 and Windows Server 2008 R2 perform similarly up to about 300,000 links per namespace. After that, Windows Server 2008 R2 performs better than Windows Server 2008. We ran the tests with up to 500,000 links per namespace.
The next chart looks into the performance of “2000 mode” domain namespace. Keep in mind that the maximum recommended number of links to “2000 mode” namespaces is 5,000 links per namespace due to limitation in which namespace data is stored in Active Directory in “2000 mode”. A subset of the “2008 mode” domain data is provided for comparison (up to 50,000 links so that you can still see the “2000 mode” data in the chart).
You can verify in the chart that the “2008 mode” domain namespace is consistently faster than the “2000 mode” domain namespace. You can also see that Windows Server 2008 and Windows Server 2008 R2 perform similarly.
Next, we look further into “2008 mode” domain namespaces, pushing the limits in terms of number of links per namespace:
You can see how Windows Server 2008 R2 scales better than Windows Server 2008 for which the performance degrades after around 300,000 links per namespace. We kept adding links in Windows Server 2008 R2 until we had 1,300,000 links, without any issues. At this point we could observe a saturation behavior similar to the one observed around 300,000 links for Windows Server 2008.
The last chart below compares the startup performance of the “2008 mode” domain namespace, with up to 200,000 links:
As you can see, Windows Server 2008 R2 consistently performs better than Windows Server 2008.
We will be presenting details on how the tests were performed and more tests results in a presentation during the upcoming Storage Developers Conference. For details, check
Post by Marcello Hasegawa