TPC-C benchmarks? Look at the most recent...

Ok, here's the article that got me started to think about this... http://www.theregister.co.uk/2009/04/09/nehalem_ep_oltp_test/ So I went to the TPC-C org's site to check out the new HP benchmark. Surprising, HP has been partnering with Oracle. Of course prior benchmarks had been with Microsoft's SQLServer. If we look at the benchmarks, Microsoft/Dell seem to be able to put out a decent platform capable of delivering 100K tpm for an average cost of $.60 USD. If you look at Oracle/HP and Microsoft/HP they are capable of putting out 630K tpm at an average cost of $1.09 per transaction. IBM? Well if you look at the IBM/Microsoft its putting out 680K tpm with an average cost of around $2.50 USD per transaction. So essentially, IBM's benchmarks on a relatively comparable basis show that any solution using IBM hardware will cost twice as much as a solution using non-ibm hardware . Of course we don't know what would have happened if we had an IDS benchmark to display. My guess is that in the realm of TPC-C IDS would pretty much blow the competition out of the water. Especially when it would be done on a Sun or even the new series of Intel's CPU. (IDS threads better than the competition.) But we won't know this because IBM doesn't release their benchmarking numbers. (Note: I said release. You can bet your bottom dollar that there are some internal unaudited benchmarks occurring. ;-) Also note: IBM hasn't done any benchmarking on Cell chips either. ;-) For those of you who also take a look at HP/Oracle and HP/Microsoft numbers, you'll see that the relative cost per transaction and the volume are very similar. So what does that tell you? -G