I took the code from my previous DataMapper examples, which was written for .Net 1.1 and SQL2K, and ported it to VS2005 Beta 1 with SQL Server 2K5. I did not have to change one line of code in either the sprocs or in the 1.1 C# code (great going VS & SQL team). Since my example code also used the DAAB 3.1, I decided to port that too, with no issues. Since I was going to be testing the beta versions running in VPC, I new that I would not be able to compare my original stats to the new stats. But since what I’m really after was performance differences versus a baseline of the DataReader, and comparing those numbers, running on a host OS versus a VPC OS should not be a factor (as long as the OS is the same, which it was, Windows XP).
So let’s look at the numbers.
The original stats (using .Net 1.1)
| DataReader |
341 |
|
| DataSet |
411 |
21% |
| Xpath |
450 |
32% |
| XmlSerialization |
542 |
59% |
Using .Net 2.0 beta 1 with SQL Server 2K5 Beta 2
| DataReader |
1545 |
|
| DataSet |
1772 |
15% |
| Xpath |
1993 |
29% |
| XmlSerialization |
2069 |
34% |
So, when comparing to a 2.0 base of DataReader, the other 3 techniques all picked up their performance (relative to the 1.1 ratios). The biggest gain was in XmlSerialization performance which cut the ratio in half (which makes sense since most of hard lifting is done in the generated code). Now I need to get down and dirty and do 2 things, tweak the .Net code to use 2.0 features (like generics and the new XPathNav) and then tweak the SQL Server sprocs to take advantage of the better SQLXML features.
But overall, considering this is still beta code and not totally performance tuned, these are incredible numbers.