HP 3000 Manuals HP 3000 Manuals
Tracing The second data collection technique HP SPT/XL uses provides information about system code or about the intrinsics called by the application. The data is collected by intercepting the calls to intrinsics made by the program, logging the necessary data, allowing the real intrinsic to be called, and again logging the required data after the intrinsic returns. This data provides information on the amount of CPU and elapsed time spent in intrinsics as well as how much time the application waited for disk I/O, locks, etc.. In some cases, certain parameters passed to the intrinsic are also logged--for example, the database name for the various database intrinsics. Logically, this tracing method can be thought of as inserting an additional library before each system library (XL.PUB.SYS and NL.PUB.SYS). These additional libraries have, for each monitored intrinsic, a procedure "stub" which logs the necessary data, calls the real intrinsic, and logs additional data after the intrinsic returns. To insert these "stubs," the process must be in an executing state. Therefore if the process is in any form of a wait state, such as a terminal read wait, HP SPT/XL will have to wait to instrument the process. If this situation arises, the user is notified with a message that allows the collection to be aborted by entering CTRL/Y if the user chooses not to wait. HP SPT/XL keeps track of the overhead (CPU and elapsed time) imposed by this tracing, and subtracts it from the total time spent in an intrinsic. Therefore, the information reported by HP SPT/XL is an accurate representation of the resources consumed by the application program. HP SPT/XL does NOT require the user to specify any additional run-time libraries when running an application. HP SPT/XL temporarily modifies the intrinsic calls to enable tracing. The intrinsics will be traced ONLY while the application is being monitored. Therefore, if the application is only monitored for 2 minutes, and continues to execute, HP SPT/XL changes the intrinsic call back to its original state when data collection ends. The application does NOT incur the overhead of tracing when monitoring is not active.