Hi all. We're working on creating operational reports to measure key metrics for Teradata 13.10. We typically use DBQL/Acctg to report on users and ResUsage tables for platform-wide metrics.
I'd like to breakdown actual CPU usage by workload definition, which is provided by the ResUsageSps table. The problem is, the sps table does not match up to the SPMA or CPU res usage tables.
CPUUServ is pretty consistent with SPMA & CPU - less than 1% error. However, CPUUExec under reports CPU by an average of 12%, ranging from 8% to 15%. If we start reporting using this table, there will be questions as to why the numbers between all of our reports are not consistent. I was wondering if anyone knows why the table under reports CPU.
We could just add 12% to all the Exec CPU reported by this table, but that is a non-ideal solution. Is this also a problem in 14.10?
If you saying that the ResUsageSPS table is reporting less CPU than the ResUsageSPMA table, this is expected behavior.
SPS only reports CPU used by the database. SPMA catches everything, so its always going to report a higher level of CPU usage. Most, but not all of what is happening on the node can be accredited to a workload directly. There is always some percent of CPU the is being consumed outside the database, by the operating system and the gateway, for example. If you have other apps running on your nodes, that would also be reported in SPMA, but not SPS table.
There is an example on page 9 of The Workload Management Capacity on Demand orange book that compares SPS and SPMA CPU usage for the same test at the same points in time and you can consisently see the difference.