We did a lot of testing for SQL Server 2000 and the Suprtdome machine was originally used for testing and benchmarkign SQL Server 2000. We have come up with many new an innovative ways to test SQL Server 2005, but it some ways what is most impressive is
not the actual testing, but the technique we use to fix the bugs, Soner and Edmund talk about this.
Remember this box does not run all the time, when it does run its very busy but the rest of the time it is relaxing so I am not sure our workload is real. I'll check wih the engineers on the disk failure thing. I do know that last time we got all new disks
for it they came on pallets that we had all over the building for storage.
I think when SQLMail came out it was a ground breaking feature, but we failed to keep it up to date in SQL Server 2000 so that has caused some of the bad experiences. We are really trying to fix that in SQL Server 20005.
I think what you are seeing with what Anders is doing and also MSR (plus some other internal projects) is us striving to modernise the way you interact with databases.
With the addition of SQLCLR into SQL Server 2005 we are providing a modern, high performance and rich extensibility model with a ton of leverage of existing skills and tools, but thats really just the start.
We now need to change the way you interact with databases, the SQl Language and its integration into the middle and client tiers has not really changed in years, I remember using VB3 Database Edition much the same way as we are now, to build client server apps.
This is an area where MSR and AndersH are doing their work.
The final area is the actual building of queries, QBE has basically not changed in 10 years, but the capabilities of the language have very much evolved and we need to allow folks to write GOOD sql who are concerned about writing it today. As a result we are
also doing work on new conceptual query builders. In fact in many ways thats one of the key features of the new reportbuilder and UDM in SQL2005. Once a technically skilled user has built the semantic layer over the database then end users can derive business
queries from that layer. We still have a lng way to go in this space but we are doing a ton of thinking.
In terms of architecture I think it is just a reflection of where we are seeing the workloads that we support heading that Soner wants to go in this direction. Remember we had no idea about adding XML, Service Broker or SQLCLR to the egnine when it was being
re-designed for SQL7 back in the 96/97 timeframe and we have been able ot add all of those in this release. I think it has evolved very effectively.
In terms of a managed database its a great question, not in the near term but maybe longer term, yes. One of the most important assets in the database is memory and its management, one of the reasons we need the Whidbey version of the CLR for hosting is that
we get more control of memory allocation and garbage collection than we do with VS2002/3, we need that in the database to make sure we get the right performance characteristics. SQL Server has its own memory, thread and disk management, we need to be able
to have that level of control to be able to truly perform.
Actually Yvonne and I worked together in Scotland before coming to work for SQL Server, as did Donald Farmer the GPM for SQL Server Integration Services (DTS). We all worked at a startup in Aberdeen and all came over at different times.