Archives of the TeradataForum

Message Posted: Mon, 20 Jan 2003 @ 22:16:13 GMT

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

A few thought to add to the discussion . . .

I find that the number of CPU seconds are diminished by Unix system processes and this accounts for approximately 20% of the total. The numbers reflected in dbc.ampusage do not account for Unix operating system processes. From there I only get about 97% of the available CPU seconds as there are some moments when the system is IO bound.

Next point to consider is that not all Teradata systems are made up of the same CPU speeds. One of mine has 3 different CPU speeds. That being the case 1 CPU second on a 700 mhz node is not the same as 1 CPU second on a 450 mhz node. With everything running at the speed of the slowest node, I prorate CPU seconds from node to node. Given that the 450 mhz nodes run at 100%, the 700 mhz nodes run at 71% of capacity. The number of CPU seconds for the 700 mhz nodes are therefore reduced by 29%.

Not sure of what configurations you have, but for the 1.4 ghz nodes with 2 CPUs they will have 1/2 the CPU seconds as does a 4 CPU node at 700 mhz. (The TPerf ratings are almost identical.)

These subtle notions come into play when dealing with coexistence systems. In reality it is only a simple spreadsheet computation.

Let me know if you have any questions on this subject.

Jim LeBlanc
NCR/Teradata Technical Account Manager

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