Archives of the TeradataForum

Message Posted: Thu, 26 Jun 2003 @ 13:50:37 GMT

	<Prev	Next>		<<First	<Prev	Next>	Last>>

To answer your questions I first need to develop the proper perspective. Thinking too much about your question leads me to contemplate the imperfection of mathematics. Thinking too lightly about the question leads me to dismiss the question outright. So with the proper balance the question has merit and I will avoid discussions such as how 1/3 expressed in decimal form is imperfect due to rounding and if rounding theory is perfect.

TPerf numbers I agree are empirical science, can mistakes be made? Of course. Observation is a human process. That is why I double check TPerf against actual experience and easily dismiss minor differences. Usually we find that with upgrades the TPerf numbers are conservative. Keep in mind that TPerf is a workload composed of CPU and IO. Different CPU and IO workloads give different results. From experience watch for those CPU intensive product joins and IO intensive workloads such as ARCMAIN to distort your mileage.

Relying upon TPerf numbers to predict system performance with 100% accuracy? Again back to the act of too closely inspecting your question, if the number is conservative then it is less than accurate in the customers' favor, minor point. So in general, I believe in the numbers and then use observation to prove the numbers.

Using TPerf as science while it is just an estimate? TPerf has yet to let me down and I check and double-check it. There have been some errors (I hear) in the TPerf calculator and then there have been some misuses of the calculator and there have been some cases when new systems were issued before the TPerf computations were done. Gosh plenty of room for people to mangle a good index.

I enjoy your question. Here is the answer to your question in summary form. Should we search for the perfect performance index (metric) where do we go how do we find that rare quality of perfection?

Thanks,

Jim LeBlanc
NCR/Teradata Technical Account Manager

	<Prev	Next>		<<First	<Prev	Next>	Last>>