Archives of the TeradataForum

Message Posted: Thu, 14 Nov 2002 @ 21:58:40 GMT

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

<-- Anonymously Posted: Thursday, November 14, 2002 16:48 -->

Macros and Views are OK. I agree that 1 second at 100% is 1 CPU second. Excellent ideas! ASE is in use and it appears that BEZ is reporting AMPUsage as the usage is reported by groups of users. Our computations expect 20 percent allocation to system. There is no intentional effort to correlate AMPUsage and ResUsage. We are just trying to reconcile why BEZ shows 80 percent and slowest node RESNode shows 98 percent. Big difference.

Yes indeed, the mix of clock speeds makes things more interesting. Calculating based on mhz alone understates capacity while calculating on observed percentages and "uninhibited TPerf" (NCR numbers) seems to be more accurate. (Use of IO inhibited TPerf also seems to incorrectly calibrate the different speeds as it includes IO performance). With all the talent this group has we'll get to the bottom of this.

Using RESUsage observed numbers, when the 450 mhz nodes are at 100 percent, the 550 mhz nodes are at 89% and the 700 mhz nodes are at 71%. Calculating this across total seconds times the number of CPUs at their respective speeds gives the total capacity for CPU seconds. These numbers are validated with uninhibited TPerf. Also when we hit this number things get hectic with performance. Also note that the number of VAMPs are identical across all nodes as are the number of disk drives and disk capacity. The only significant factor out of balance is CPU speed.

The real indicator is that the slow nodes are at 98% all day. With performance being limited by the slowest node it would seem reasonable that CPU utilization for the entire system would be 98%. But BEZ reports at least give the visual impression that the system is running at about 80 percent.

Does BEZ take the average of the mixture of all the CPUs and in doing so understate usage? I wonder . . .

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