On emulated MCP systems and MCP systems that do not support software-based capacity on demand (COD), all processors on a system run at the same processor speed. Only one set of times is used to fine-tune your system and provide billing data. These times are referred as “effective” times and consist of the task process time, initial p-bit process time, and other p-bit process time.
On MCP systems that support software-based COD, processor performance is a dynamic process. That is, COD keys and licensed partition characteristics (partition images) can be dynamically changed. Metering introduces the concept of a governor in which a system administrator has direct control of processor performance levels.
The introduction of Java processors has further changed COD by defining processor sets within a partition, each of which can run at a different performance level. Thus, a unified, extensible approach has been introduced to systemwide and task process time accounting.
Task Times
Task times have been grouped into several time values to suit the requirements of various applications.
-
Effective process time
This task time is associated with traditional processor time in which the value of a processor second varies as the processor performance level changes. Effective time values are available for task process time, initial p-bit process time, and other p-bit process time.
-
Normalized process time
This task time reflects the processing resources used by a task in units of a “standard CPU second,” which are independent of the actual processor performance level and are useful for reflecting billable resource usage. Normalized time values are available for task process time, initial p-bit process time, and other p-bit process time.
-
Measured process time
This task time reflects the actual time measured by the MCP. Depending on the processor throttling mechanism used by the MCP, this time might include the processor throttling time component when measuring processor time statistics. However, unlike effective process times and normalized process times that might be estimated time values, measured process time always reflects the measurements gathered for systemwide processor time accounting.
Thus, measured process time is useful in applications that determine percentages of systemwide resources used. Measured time values are available for task process time, initial p-bit process time, and other p-bit process time.
Task Time Comparison
A comparison of the various task times follows.
-
The effective process time represents the time that the task appears to have been on the processor (the time used for displaying commands such as ?TI and ODT ACTIVE). This time continues to be used as before.
-
The normalized process time represents a billable entity that uses a standard CPU second as the time unit.
-
The measured process time represents the time the task was actually on the processor as measured by the MCP.
Measured process time of all tasks on the system since a halt/load represents the time measured in the systemwide time statistics.
-
The sum of all task-measured process times since a halt/load is represented by the sum of systemwide USER STACK counters plus systemwide MCP STACK counters across all processor sets.
-
The sum of all task-measured initial p-bit process times since a halt/load is represented by the sum of systemwide INITIAL PBIT counters across all processor sets.
-
The sum of all task-measured other p-bit process times since a halt/load is represented by the sum of systemwide OTHER PBIT counters across all processor sets.
More Information
For more information about interfaces that return systemwide processor time statistics and task processor time statistics, refer to the following manuals.
|
In the . . . |
Refer to . . . |
|---|---|
|
SYSTEMSTATUS Programming Reference Manual |
Utilization Accounts (Type 25) |
|
System Log Programming Reference Manual |
|
|
ALGOL Programming Reference Manual, Volume 1: Basic Implementation |
TIME Intrinsic |
|
Task Attributes Programming Reference Manual |
|
