Summary

Overall, the testing showed SQL Server 2005 is an improvement over SQL Server 2000 in ease of use and provides great performance gains with no additional tuning to existing code. The initial graphs show more disk activity and longer disk queues with shorter run times, which leads one to believe that 2005 has more parallelism and/or more asynchronous I/O - both good for throughput. These graphs also show more CPU utilization which, in light of the shorter run times, means more work is being completed per unit of time via parallelism/asynchronous I/O and/or improved execution efficiency (code tuning/compiler options, better memory utilization, or other means like improved locking). This is usually good overall because it means shorter batch run times. On the other hand, the increased CPU can also mean more pressure on storage, and in a mixed batch/online environment could require a re-tuning to throttle back batch CPU or storage utilization so online transactions can keep its original (SQL 2000) share of resources and original response times. It also looks like there are some areas where noticeably less CPU is used for a given task which is also a favorable indicator for batch and online.

Additionally, the data shows a few cases where the run times are longer with SQL2005, but that often happens with complex software where improvements in one area hurt another. However, SQL2005 relational queries showed much improved response times when compared to SQL2000. Perhaps as a result of query optimizer and tuning enhancements / overall better overall resource utilization by the relational engine.

Overall, these optimistic proof points demand more testing and investigation into SQL 2005 with a focus on the adoption of SQL 2005 T-SQL enhancements to induce performance gains in ETL processing similar to those seen with relational queries. OLAP and Analysis Server Cube generation seem to be the next logical functional steps. The testing provided Raymond James an invaluable gauge of the current state of SQL 2005 as well as a testimony of teaming of RJ, Microsoft, Intel, and HP. These initial results provide a clear roadmap and direction for further investigation.

Needless to say, with these results, Raymond James has put a plan in place to upgrade their enterprise to SQL Server 2005. We would like to acknowledge Bruce Phillipoom and Chad Miller of Raymond James Financial, Randy Baklini and the entire HP team for allowing us to share this great work with you.