Why are CPU EFFECTIVENESS percentages on z990 hardware sometimes slightly over 100%
For a full description of CPU EFFECTIVENESS see knowledge base article TEC550739.
On z990 hardware, it is possible for the IDMS CPU Effectiveness value to show a slightly higher value than what is actually being achieved.
The reason for this is on z990 hardware the instruction which is required to capture the Elapsed time is the "STCK" instruction.
The "STCK" instruction requires CPU cycles to coordinate a value across all processors on the "LPAR".
These additional cycles increase relative to the number of processors on the LPAR, in other words, the CPU cycles required for the coordination are significantly less with a smaller number of processors, the CPU cycles required do increase as the number of processors on the LPAR increase.
The STCK instruction is designed to guarantee a unique time stamp across all CPUs in an LPAR and if the LPAR is part of a sysplex then across the entire sysplex. Whenever possible IDMS uses STCKF which merely returns the TOD value without attempting to guarantee uniqueness, unfortunately a z990 does not support the STCKF instruction and therefore the STCK is used.
One of the results of this can be the TIMEUSED on an IDMS system could actually show higher than the elapsed time reported.
This in turn would result in a CPU effectiveness value which exceeds 100 percent as shown by DCMT D SUBT EFF, the values reported would normally be no higher than 102 or 103 percent. Values exceeding 100 percent would only be shown on IDMS systems which are getting all of the CPU cycles they required while they are dispatched within the operating system.
On the same hand, given the z990 with a high number of processors on the LPAR, a reported CPU effectiveness value lower than 100 percent could be considered to be slightly over reported.
This situation does not exist on z196 hardware as the STCKF instruction is used which is more efficient than the "STCK" instruction.
Release: IDADSO00100-18.5-ADS-for CA-IDMS