In software engineering, performance analysis, more commonly today known as profiling, is the investigation of a program's behavior using information gathered as the program executes (i.e. it is a form of dynamic program analysis, as opposed to static code analysis). The usual goal of performance analysis is to determine which sections of a program to optimize - usually either to increase its speed or decrease its memory requirement (or sometimes both).

Software Testing portal

Contents

[edit] Use of profilers

A profiler is a performance analysis tool that measures the behavior of a program as it executes, particularly the frequency and duration of function calls. (An instruction set simulator which is also - by necessity - a profiler, can measure the totality of a programs behaviour from invocation to termination.) The output may be a stream of recorded events (a trace) or a statistical summary of the events observed (a profile) or an ongoing interaction with the hypervisor. Profilers use a wide variety of techniques to collect data, including hardware interrupts, code instrumentation, instruction set simulation, operating system hooks, and performance counters. The usage of profilers is 'called out' in the performance engineering process.

As the summation in a profile often is done related to the source code positions where the events happen, the size of measurement data is linear to the code size of the program. In contrast, the size of a (full) trace is linear to the program's execution time, making it somewhat impractical. For sequential programs, a profile is usually enough, but performance problems in parallel programs (waiting for messages or synchronization issues) often depend on the time relationship of events, thus requiring the full trace to get an understanding of the problem.

Program analysis tools are extremely important for understanding program behavior. Computer architects need such tools to evaluate how well programs will perform on new architectures. Software writers need tools to analyze their programs and identify critical sections of code. Compiler writers often use such tools to find out how well their instruction scheduling or branch prediction algorithm is performing... (ATOM, PLDI, '94)

[edit] History

Performance analysis tools existed on IBM/360 and IBM/370 platforms from the early 1970s, usually based on timer interrupts which recorded the Program status word (PSW) at set timer intervals to detect "hot spots" in executing code. This was an early example of sampling (see below). In early 1974, Instruction Set Simulators permitted full trace and other performance monitoring features.

Profiler-driven program analysis on Unix dates back to at least 1979, when Unix systems included a basic tool "prof" that listed each function and how much of program execution time it used. In 1982, gprof extended the concept to a complete call graph analysis (Gprof: a Call Graph Execution Profiler [1])

In 1994, Amitabh Srivastava and Alan Eustace of Digital Equipment Corporation published a paper describing ATOM [2]. ATOM is a platform for converting a program into its own profiler. That is, at compile time, it inserts code into the program to be analyzed. That inserted code outputs analysis data. This technique, modifying a program to analyze itself, is known as "instrumentation".

In 2004, both the Gprof and ATOM papers appeared on the list of the 50 most influential PLDI papers of all time. [3]

[edit] Profiler Types based on Output

[edit] Flat profiler

Flat profilers compute the average call times, from the calls, and do not breakdown the call times based on the callee or the context.

[edit] Call-Graph profiler

Call Graph profilers show the call times, and frequencies of the functions, and also the call-chains involved based on the callee. However context is not preserved.

[edit] Methods of data gathering

[edit] Event based profilers

The programming languages listed here have event-based profilers:

  1. .NET: Can attach a profiling agent as a COM server to the CLR. Like Java, the runtime then provides various callbacks into the agent, for trapping events like method JIT / enter / leave, object creation, etc. Particularly powerful in that the profiling agent can rewrite the target application's bytecode in arbitrary ways.
  2. Java: JVM-Tools Interface (formerly JVM Profiling Interface) JVM API provides hooks to profiler, for trapping events like calls, class-load, unload, thread enter leave.
  3. Python: Python profiling includes the profile module, hotshot (which is call-graph based), and using the 'sys.setprofile()' module to trap events like c_{call,return,exception}, python_{call,return,exception}.
  4. Ruby: Ruby also uses a similar interface like Python for profiling. Flat-profiler in profile.rb, module, and ruby-prof a C-extension are present.

[edit] Statistical profilers

Some profilers operate by sampling. A sampling profiler probes the target program's program counter at regular intervals using operating system interrupts. Sampling profiles are typically less accurate and specific, but allow the target program to run at near full speed.

Some profilers instrument the target program with additional instructions to collect the required information. Instrumenting the program can cause changes in the performance of the program, causing inaccurate results and heisenbugs. Instrumenting can potentially be very specific but slows down the target program as more specific information is collected.

The resulting data are not exact, but a statistical approximation. The actual amount of error is usually more than one sampling period. In fact, if a value is n times the sampling period, the expected error in it is the square-root of n sampling periods. [4]

Some of the most commonly used statistical profilers are GNU's gprof, Oprofile, AMD's CodeAnalyst and SGI's Pixie.

[edit] Instrumentation

  • Manual: Done by the programmer, e.g. by adding instructions to explicitly calculate runtimes.
  • Compiler assisted: Example: "gcc -pg ..." for gprof, "quantify g++ ..." for Quantify
  • Binary translation: The tool adds instrumentation to a compiled binary. Example: ATOM
  • Runtime instrumentation: Directly before execution the code is instrumented. The program run is fully supervised and controlled by the tool. Examples: PIN, Valgrind
  • Runtime injection: More lightweight than runtime instrumentation. Code is modified at runtime to have jumps to helper functions. Example: DynInst

[edit] Hypervisor/Simulator

[edit] Simple manual technique

A variation on the sampling technique, based on deep sampling, trades accuracy for efficacy. It requires no instrumentation and takes only a small number of random time samples (e.g. 20). This can be done by running the program under a debugger and interrupting it manually at unpredictable times using a Break key or Esc key. (These samples are taken during the time when the program is being subjectively slow. If needed, a temporary outer loop can be added, to make it run long enough to take manual samples.) At each sample, the entire call stack consisting of the program counter plus the stack of return addresses is recorded. Then the call stack samples are examined for addresses that appear on multiple samples.

Any address that is on the call stack X% of the time identifies an instruction which, if removed, will save X% of running time.

For example, suppose a program contains (somewhere) a function call instruction or statement that could in principle be worked around or removed, and suppose it accounts for X = 60% of execution time. It doesn't matter whether this is distributed over one long invocation or many short ones. The instruction will appear on 60% of samples, more or less. The larger the percentage is, the fewer samples are needed to display the instruction. (In the limiting case of an infinite loop, only one sample is needed.) No guesswork is required other than the need to ask which instructions that appear on multiple samples could be removed. Such removal may require revision of code or data structure. When this is done, a performance factor of up to 1/(1-X) is gained. In this case 1/(1-0.6) = 2.5 times.

(Recursion is not an issue, because even if an instruction occurs multiple times in one call stack sample, it still only counts as one sample.)

Experience shows that this entire process can be repeated multiple times, up to a point of diminishing returns, often resulting in large cumulative speedups.

[edit] See also

[edit] References

  • Dunlavey, “Performance tuning with instruction-level cost derived from call-stack sampling”, ACM SIGPLAN Notices 42, 8 (August, 2007), pp. 4-8.
  • Dunlavey, “Performance Tuning: Slugging It Out!”, Dr. Dobb's Journal, Vol 18, #12, November 1993, pp 18-26.

[edit] External links