HP 3000 Manuals HP 3000 Manuals
Sampling The first data collection technique provides information about where, within application code, most of the time is being spent. This collection technique is often referred to as "sampling." To accomplish this requires frequent logging (sampling) of the location of the instruction the application is currently executing. By sampling the application often enough, a statistical representation of where the application is executing can be obtained. For example, if an application is sampled every few milliseconds for a few minutes, the data can be combined to determine exactly what procedures were executing most often and therefore where most of the time was spent. Since it is common for an application to spend most of its time in system code, the profile produced from this sampling data often highlights the internal operating system routines that are executed when system intrinsics are called. HP SPT/XL resolves this problem by disabling sampling while intrinsics execute and re-enabling sampling upon returning to user code. This allows HP SPT/XL to provide a more representative profile of where user code actually executes. The system intrinsics are profiled using a tracing data collection technique described below. The actual technique HP SPT/XL uses to sample the user application is different depending on the version of HP SPT/XL being used. Versions that execute on pre-MPE XL 3.0 releases use a sampling technique based on a time interval. That is, the location of the instruction the application is executing is sampled every few milliseconds. The sampling interval that HP SPT/XL uses depends on the model of HP3000 on which the data collection is done. The sampling interval varies between models because different models of HP3000 execute different numbers of instructions in a given time. The sampling rate will include approximately the same number of executed instructions on all models of the HP3000. On versions of HP SPT/XL running on MPE XL 3.0 and later releases, the sampling interval is based directly on the number of instructions executed. That is, the location where the application is executing is sampled after so many instructions have been executed. The sampling rate is approximately the same as with the time-based sampling technique. No matter what sampling technique is used, after data collection has ended HP SPT/XL logs the names and addresses of the procedures in the program file and in the executable libraries used by the application. HP SPT/XL uses these names and addresses to cluster the samples into the various procedures used, and present a profile--based on procedure name--of where time was spent in the application. By default, HP SPT/XL does NOT log the procedure names for the two system libraries (XL.PUB.SYS and NL.PUB.SYS), and a sampling profile by procedure name is not available for these libraries. At the time of collection, an option is available to inform HP SPT/XL to log this data. This collection option is called the SYSPROCNAMES option. If this option is enabled, about 1-3 megabytes of procedure name data will be placed in the logfile after data collection has ended. See the HP SPT/XL User's Manual:Collection Software for more details on the SYSPROCNAMES option.