<?xml version="1.0" encoding="latin1" ?>
<!DOCTYPE chapter SYSTEM "chapter.dtd">

<chapter>
  <header>
    <copyright>
      <year>2001</year><year>2011</year>
      <holder>Ericsson AB. All Rights Reserved.</holder>
    </copyright>
    <legalnotice>
      The contents of this file are subject to the Erlang Public License,
      Version 1.1, (the "License"); you may not use this file except in
      compliance with the License. You should have received a copy of the
      Erlang Public License along with this software. If not, it can be
      retrieved online at http://www.erlang.org/.
    
      Software distributed under the License is distributed on an "AS IS"
      basis, WITHOUT WARRANTY OF ANY KIND, either express or implied. See
      the License for the specific language governing rights and limitations
      under the License.
    
    </legalnotice>

    <title>Profiling</title>
    <prepared>Ingela Anderton</prepared>
    <docno></docno>
    <date>2001-11-02</date>
    <rev></rev>
    <file>profiling.xml</file>
  </header>

  <section>
    <title>Do not guess about performance - profile</title>

    <p>Even experienced software developers often guess wrong about where
    the performance bottlenecks are in their programs.</p>

    <p>Therefore, profile your program to see where the performance
    bottlenecks are and concentrate on optimizing them.</p>

    <p>Erlang/OTP contains several tools to help finding bottlenecks.</p>

    <p><c>fprof</c> and <c>eprof</c> provide the most detailed information
    about where the time is spent, but they significantly slow downs the
    programs they profile.</p>

    <p>If the program is too big to be profiled by <c>fprof</c> or <c>eprof</c>,
    <c>cover</c> and <c>cprof</c> could be used to locate parts of the
    code that should be more thoroughly profiled using <c>fprof</c> or
    <c>eprof</c>.</p>

    <p><c>cover</c> provides execution counts per line per process,
    with less overhead than <c>fprof/eprof</c>. Execution counts can
    with some caution be used to locate potential performance bottlenecks.
    The most lightweight tool is <c>cprof</c>, but it only provides execution
    counts on a function basis (for all processes, not per process).</p>
  </section>

  <section>
    <title>Big systems</title>
    <p>If you have a big system it might be interesting to run profiling
      on a simulated and limited scenario to start with. But bottlenecks
      have a tendency to only appear or cause problems when
      there are many things going on at the same time, and when there
      are many nodes involved. Therefore it is desirable to also run
      profiling in a system test plant on a real target system.</p>
    <p>When your system is big you do not want to run the profiling
      tools on the whole system. You want to concentrate on processes
      and modules that you know are central and stand for a big part of the
      execution.</p>
  </section>

  <section>
    <title>What to look for</title>
    <p>When analyzing the result file from the profiling activity
      you should look for functions that are called many
      times and have a long "own" execution time (time excluding calls
      to other functions). Functions that just are called very
      many times can also be interesting, as even small things can add
      up to quite a bit if they are repeated often. Then you need to
      ask yourself what can I do to reduce this time. Appropriate
      types of questions to ask yourself are: </p>
    <list type="bulleted">
      <item>Can I reduce the number of times the function is called?</item>
      <item>Are there tests that can be run less often if I change
       the order of tests?</item>
      <item>Are there redundant tests that can be removed? </item>
      <item>Is there some expression calculated giving the same result
       each time? </item>
      <item>Are there other ways of doing this that are equivalent and
       more efficient?</item>
      <item>Can I use another internal data representation to make
       things more efficient? </item>
    </list>
    <p>These questions are not always trivial to answer. You might
      need to do some benchmarks to back up your theory, to avoid
      making things slower if your theory is wrong. See <seealso marker="#benchmark">benchmarking</seealso>.</p>
  </section>

  <section>
    <title>Tools</title>

    <section>
      <title>fprof</title>
      <p><c>fprof</c> measures the execution time for each function,
        both own time i.e how much time a function has used for its
        own execution, and accumulated time i.e. including called
        functions. The values are displayed per process. You also get
        to know how many times each function has been
        called. <c>fprof</c> is based on trace to file in order to
        minimize runtime performance impact. Using fprof is just a
        matter of calling a few library functions, see fprof manual
        page under the application tools.</p>
      <p><c>fprof</c> was introduced in version R8 of Erlang/OTP. Its
        predecessor <c>eprof</c> that is based on the Erlang trace BIFs,
        is still available, see eprof manual page under the
        application tools. Eprof shows how much time has been used by
        each process, and in which function calls this time has been
        spent.  Time is shown as percentage of total time, not as
        absolute time.</p>
    </section>

    <section>
      <title>cover</title>
      <p><c>cover</c>'s primary use is coverage analysis to verify
        test cases, making sure all relevant code is covered.
        <c>cover</c> counts how many times each executable line of
        code is executed when a program is run. This is done on a per
        module basis. Of course this information can be used to
        determine what code is run very frequently and could therefore
        be subject for optimization. Using cover is just a matter of
        calling a few library functions, see cover manual
        page under the application tools.</p>
    </section>

    <section>
      <title>cprof</title>
      <p><c>cprof</c> is something in between <c>fprof</c> and
        <c>cover</c> regarding features. It counts how many times each
        function is called when the program is run, on a per module
        basis. <c>cprof</c> has a low performance degradation effect (versus
        <c>fprof</c> and <c>eprof</c>) and does not need to recompile
        any modules to profile (versus <c>cover</c>).</p>
    </section>

    <section>
      <title>Tool summarization</title>
      <table>
        <row>
          <cell align="center" valign="middle">Tool</cell>
          <cell align="center" valign="middle">Results</cell>
          <cell align="center" valign="middle">Size of result</cell>
          <cell align="center" valign="middle">Effects on program execution time</cell>
          <cell align="center" valign="middle">Records number of calls</cell>
          <cell align="center" valign="middle">Records Execution time</cell>
          <cell align="center" valign="middle">Records called by</cell>
          <cell align="center" valign="middle">Records garbage collection</cell>
        </row>
        <row>
          <cell align="left" valign="middle"><c>fprof </c></cell>
          <cell align="left" valign="middle">per process to screen/file </cell>
          <cell align="left" valign="middle">large </cell>
          <cell align="left" valign="middle">significant slowdown </cell>
          <cell align="left" valign="middle">yes  </cell>
          <cell align="left" valign="middle">total and own</cell>
          <cell align="left" valign="middle">yes </cell>
          <cell align="left" valign="middle">yes </cell>
        </row>
        <row>
          <cell align="left" valign="middle"><c>eprof </c></cell>
          <cell align="left" valign="middle">per process/function to screen/file </cell>
          <cell align="left" valign="middle">medium </cell>
          <cell align="left" valign="middle">significant slowdown </cell>
          <cell align="left" valign="middle">yes </cell>
          <cell align="left" valign="middle">only total </cell>
          <cell align="left" valign="middle">no </cell>
          <cell align="left" valign="middle">no </cell>
        </row>
        <row>
          <cell align="left" valign="middle"><c>cover </c></cell>
          <cell align="left" valign="middle">per module to screen/file</cell>
          <cell align="left" valign="middle">small </cell>
          <cell align="left" valign="middle">moderate slowdown</cell>
          <cell align="left" valign="middle">yes, per line  </cell>
          <cell align="left" valign="middle">no </cell>
          <cell align="left" valign="middle">no </cell>
          <cell align="left" valign="middle">no </cell>
        </row>
        <row>
          <cell align="left" valign="middle"><c>cprof </c></cell>
          <cell align="left" valign="middle">per module to caller</cell>
          <cell align="left" valign="middle">small </cell>
          <cell align="left" valign="middle">small slowdown </cell>
          <cell align="left" valign="middle">yes </cell>
          <cell align="left" valign="middle">no </cell>
          <cell align="left" valign="middle">no </cell>
          <cell align="left" valign="middle">no </cell>
        </row>
        <tcaption></tcaption>
      </table>
    </section>
  </section>

  <section>
    <marker id="benchmark"></marker>
    <title>Benchmarking</title>

    <p>The main purpose of benchmarking is to find out which
    implementation of a given algorithm or function is the fastest.
    Benchmarking is far from an exact science. Today's operating systems
    generally run background tasks that are difficult to turn off.
    Caches and multiple CPU cores doesn't make it any easier.
    It would be best to run Unix-computers in single-user mode when
    benchmarking, but that is inconvenient to say the least for casual
    testing.</p>
    
    <p>Benchmarks can measure wall-clock time or CPU time.</p>

    <p><seealso marker="stdlib:timer#tc/3">timer:tc/3</seealso> measures
    wall-clock time. The advantage with wall-clock time is that I/O,
    swapping, and other activities in the operating-system kernel are
    included in the measurements. The disadvantage is that the
    the measurements will vary wildly. Usually it is best to run the
    benchmark several times and note the shortest time - that time should
    be the minimum time that is possible to achieve under the best of
    circumstances.</p>

    <p><seealso marker="erts:erlang#statistics/1">statistics/1</seealso>
    with the argument <c>runtime</c> measures CPU time spent in the Erlang
    virtual machine. The advantage is that the results are more
    consistent from run to run. The disadvantage is that the time
    spent in the operating system kernel (such as swapping and I/O)
    are not included. Therefore, measuring CPU time is misleading if
    any I/O (file or socket) is involved.</p>

    <p>It is probably a good idea to do both wall-clock measurements and
    CPU time measurements.</p>

    <p>Some additional advice:</p>

    <list type="bulleted">
    <item>The granularity of both types of measurement could be quite
    high so you should make sure that each individual measurement
    lasts for at least several seconds.</item>

    <item>To make the test fair, each new test run should run in its own,
    newly created Erlang process. Otherwise, if all tests run in the
    same process, the later tests would start out with larger heap sizes
    and therefore probably do less garbage collections. You could
    also consider restarting the Erlang emulator between each test.</item>

    <item>Do not assume that the fastest implementation of a given algorithm
    on computer architecture X also is the fastest on computer architecture Y.</item>

    </list>
  </section>
</chapter>