Tech Analysis: Windows Vista Sucks Performance

After running PerformanceTest under Vista, engineers installed and ran Microsoft's XML Mark 11 test to measure the .Net runtime performance. They then did the same with XP. The XML Mark 11 test can be downloaded from Microsoft's Visual Studio Download Center. XML Mark 11 simulates a multi-threaded server application running a SAX and a DOM parser. The test measures number of threads, retrieval speed of items in an XML document and various transaction processes. In addition to .Net, the XML Mark includes Java code. However, engineers used the C# portion of the test without modifying the code.

The XML Mark test was originally written by Sun Microsystems about two years ago to show Java's XML processing speed relative to .Net 1.1's XML parsing. With the release of .Net 2.0, however, Microsoft adopted the test and was able to show that C# code ran just as fast as Java code.

Because Vista includes .Net 3.0, engineers had to recompile the C# code using Visual Studio 2005 extensions for .NET Framework 3.0 (WCF & WPF), from the November 2006 CTP version. Microsoft does not have a finished .Net 3.0 framework out yet for Visual Studio developers. In addition, engineers had to install the .Net Framework 3.0 Redistributable package on XP.

During compilation, a few warning messages were received because a couple of the XML API methods were obsolete. Microsoft changed some of the XML parsing technologies in its .Net framework between versions 2.0 and 3.0. For XML Mark to compile and run properly on both operating systems, engineers added .Net 3.0 as an additional prerequisite to .Net 2.0, instead of eliminating .Net 2.0 from the compilation process completely.

As suspected, Vista hampered .Net's performance as well.

The XML Mark showed that XML applications run faster on XP than on Vista when both operating systems use .Net 3.0 runtime. However, the test produced different results between the two parsers. With the DOM parser test, XP ran about 20 percent faster than Vista. However, Vista ran about 3 percent slower with the stream parser (SAX) test.

These results were expected and were in line with the PerformanceTest results. The DOM test provided a more precise view of .Net's memory architecture because of the large objects it created and deleted. .Net's garbage collection was far more stressed with the DOM test than with the stream test.

The results show that Vista's memory architecture is significantly slower than XP. Microsoft released a product that will make many .Net applications run slower. At this point, we do not recommend running client-based .Net applications on Vista that require large data processing. And unless it's imperative that users have an operating system with a more exciting look and feel, XP will offer better performance than Vista.