Windows Server 2008 and SQL Server 2008 Benchmarks

Just think, if I publish a TPC-H on my laptop, I'd get a top 10 ranking.  ;-)

Great Info!

That's good because it means that SQL2008 will be released in six months (in August). Here's the better context in which to view the TPC-H result: http://www.tpc.org/tpch/results/tpch_perf_results.asp.

Sorry, why do you think it'll be August? Has the release date been shifted?

Because the TPC benchmarks mandate that the system must be available in six months of the publication of any official benchmark result. Otherwise, the benchmark result must be withdrawn. I just checked, and the benchmark results indicate August 30 as the availability date.

OK, that's nice for the BI\DW side of the house.  Now impress me on the OLTP side with either TPC-C or E.  

So far, the only comparitive postings at this time are from IBM on the TPC-E, and there a 12% out of the box performance improvement is observed.

Linchi's right about the TPC's six-month rule. To my knowledge, SQL has never had to withdraw a result due to this rule, but I'm sure we'd do that if it were necessary to ship a high-quality final product. That we were able to deliver fairly diverse benchmark results, mostly on the February CTP bits, is a positive sign.

Larry: In addition to the new IBM x-Series TPC-E result, which shows a nice gain over SQL Server 2005, NEC also published SQL's first 64-core Itanium result on TPC-E. This is the new #1 TPC-E result at 1126 tpsE.

For OLTP, you won't see us publishing TPC-C results any longer, starting with SQL Server 2008; TPC-E is our new OLTP lighthouse. TPC-E is a far better model of modern OLTP applications, with referential integrity checks, realistic synthetic data, requirements for reliable storage, and a richer schema that has far more FK relationships than TPC-C, and uses a broader set of data types. Our experience with TPC-E has shown that customers are more likely to see gains on TPC-E benchmark results reflected on their own servers running their own OLTP apps.

> For OLTP, you won't see us publishing TPC-C results any longer

I know MS has turned its attention to TPC-E for whatever reason. But not publishing TPC-C at all? I just checked TPC-C results, and the most recent TPC-C result from MS was as recent as Jan 7, 2008 with SQL2005.

Microsoft just release a TPC-C on Dunnington 4 socket hexa core on SQL Server 2005 on Aug 19th 2008!! (http://www.tpc.org/tpcc/results/tpcc_result_detail.asp?id=108081901)

So this goes against what's being said here and would lead some folks to feel that perhaps SS08 is purposely tuned for TPC-E and that hurts TPC-C? Perhaps consipracy theory, but odd indeed.

I don't disagree that TPC-E is more up to date, but so long as only SQL Server does it, results are moot. The market decides what is up to date and that seems to be TPC-C at the same time. To add that MS has had a large scale TPC-C result in a long time and likely couldn't topple today's leader (or ever last years) I think is a pretty big reason (mind you the top spots are theory benchmarks since nobody does that level of activity).

Let me close on why I like TPC-C. In the full disclosure you have to show the logger -- and all thing being =, in high end OLTP this is the chokepoint. When SQL Server logs about 6 KB/TX and Oracle and DB2 log less , it tells me there is a logging issue. When each time SQL Server checkpoints, performance drops, that is an issue. That's why I care about TPC-C -- it shows me how the database performs and the details behind it; not because I think it represents the real world.