glossary

A state of a procedure. Puma considers a procedure to be active if control is in the procedure itself or in any of its descendants.

A procedure that calls the given procedure, either directly or through another procedure or procedures. Contrast with parent.

A procedure called directly by the given procedure. Contrast with descendant.

An analysis specification that instructs Puma to perform recursion collapsing unless doing so would omit any routine names out of the call chain.

The collective name for Puma's target program, the directory where the target program executes, and the pathname of the file to which Puma writes the collected data. Puma context also includes any search directories you have provided for target source files.

A Puma analysis mechanism that produces a raw count (as opposed to a percentage) for a given statistic in the analysis.

A file in which Puma stores the data that it collects from a process. Data files are used during analysis and playback.

A procedure that is called by the given procedure, either directly or through other procedures. Contrast with child.

An analysis specification that instructs Puma to perform recursion collapsing only for direct recursion (that is, when a routine calls itself directly).

A Puma analysis that indicates the overall percentage of time spent in-only and/or in-or-under each procedure, regardless of the procedure's ancestors. A flat report does not reflect the dynamic calling structure of a program. Contrast with Hierarchical report.

See Program flow.

An analysis specification that instructs Puma to perform recursion collapsing whenever it encounters a routine for which there is a higher instance in the call tree.

The degree of detail in which Puma displays analysis data. Data may be shown separately for each routine or library of a program; or data may be aggregated and associated with the top-level caller. The former granularity is called "in-only"; the latter is called "in-or-under."

A Puma analysis that reflects the dynamic calling structure of a program, giving a structured breakdown of where the program spends its time. Contrast with Flat report.

A term used to refer to execution time spent within a procedure but not within its descendants. Contrast with in-or-under.

A term used to refer to execution time spent within a procedure and/or its descendants. Contrast with in-only.

See sample interval.

A specification that limits the nesting depth of a Puma hierarchical analysis to a certain number of levels. The default is a level cutoff of 64.

As a Puma data-collection option, the term library refers to a dynamically loaded library; that is, a collection of executable routines bound together. Library routines are loaded into the target program when one of the routines is called by the target program.

The lowest percentage of execution time included in a Puma display. The default limit is 1.

To collect performance data from an executing program.

An analysis specification that instructs Puma to collapse none of the recursion of a program, but instead to show the full dynamic tree.

The time spent by Puma recording and displaying a sample during data collection.

A procedure that directly calls the given procedure. Contrast with ancestor.

A specification that restricts entries in a Puma analysis to procedures consuming more than a certain amount of execution time. By default, the percent cutoff is 0, which, in effect, instructs Puma to include information about every procedure that consumes 0% or more of the execution time (that is, every procedure that occurred in any stacktrace that Puma recorded).

A Puma analysis mechanism that produces a statistical value as a percentage of the value for the whole set of samples being analyzed.

A set of metrics that Puma collects from a program run to aid in analyzing performance. These include time spent in user space ("user cycles"), time spent in system space ("system cycles"), major page faults, minor page faults, socket messages sent, socket messages received, signals received, files open, page swaps, and characters (bytes) of I/O.

To use data accumulated while monitoring a program to examine the sequence of the execution of the program.

A cutoff that excludes calls beneath a given procedure from the analysis.

A binary executed by the CPU. Each process is executed independently.

A graphical image of the currently active stack trace and the siblings of each procedure in that stack trace.

A register in the CPU that contains the address of the next instruction to be executed.

The procedure path through which the execution of a program has gone to bring the program counter to its current location.

Any situation in which procedure calls are circular; for example, when a procedure calls itself, when a procedure calls the procedure that called it, or when a procedure calls farther back up the call chain.

A Puma analysis mechanism for reporting on the recursion of a program. When you request recursion collapsing, Puma may not make a new node in a tree of call chains if it encounters a recursive call (depending on the type of recursion collapsing you specify). Instead Puma creates a recursive stub that represents the recursion and refers to the place higher up the call chain where the same routine occurs. Puma then jumps up in the tree to that point and continues playing out the stack trace.

See recursion collapsing.

See Performance statistics.

The data obtained from one interruption of an executing program. The sample includes statistical information and a stack trace.

The period of time between samples taken during data collection.

The number of samples Puma gathers per second that a target program has been allowed to run during data collection.

A procedure with the same parent as the given procedure.

The sequence of procedure calls leading to the program counter at any given time.

To reconstruct the execution of a program one sample at a time.

A program from which Puma collects performance data.

An independent stream of program execution. A program may be made up of one or more threads; multiple threads in a program cooperate in solving a common problem.

The time the processing unit spends in the user's program, excluding time spent waiting for I/O or executing other programs.

The total time taken for program execution, including disk and memory accesses, I/O, and operating system overhead.