path: root/lib/tools/doc/src/fprof.xml

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

<erlref>
  <header>
    <copyright>
      <year>2001</year><year>2009</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>fprof</title>
    <prepared>Raimo Niskanen</prepared>
    <responsible>nobody</responsible>
    <docno></docno>
    <approved>nobody</approved>
    <checked></checked>
    <date>2001-08-13</date>
    <rev>PA1</rev>
    <file>fprof.sgml</file>
  </header>
  <module>fprof</module>
  <modulesummary>A Time Profiling Tool using trace to file for minimal runtime performance impact.</modulesummary>
  <description>
    <p>This module is used to profile a program
      to find out how the execution time is used.
      Trace to file is used to minimize 
      runtime performance impact. 
      </p>
    <p>The <c>fprof</c> module uses tracing to collect profiling data, 
      hence there is no need for special compilation of any module to
      be profiled. When it starts tracing, <c>fprof</c> will erase all 
      previous tracing in the node and set the necessary trace flags
      on the profiling target processes as well as local call trace on
      all functions in all loaded modules and all modules to be loaded.
      <c>fprof</c> erases all tracing in the node when it stops tracing.
      </p>
    <p><c>fprof</c> presents both <em>own time</em> i.e how much time a
      function has used for its own execution, and 
      <em>accumulated time</em> i.e including called functions. 
      All presented times are
      collected using trace timestamps. <c>fprof</c> tries to collect
      cpu time timestamps, if the host machine OS supports it. 
      Therefore the times may be wallclock times and OS scheduling will 
      randomly strike all called functions in a presumably fair way.
      </p>
    <p>If, however, the profiling time is short, and the host machine
      OS does not support high resolution cpu time measurements, some
      few OS schedulings may show up as ridiculously long execution
      times for functions doing practically nothing. An example of a
      function more or less just composing a tuple in about 100 times
      the normal execution time has been seen, and when the tracing
      was repeated, the execution time became normal.
      </p>
    <p>Profiling is essentially done in 3 steps:</p>
    <taglist>
      <tag><c>1</c></tag>
      <item>Tracing; to file, as mentioned in the previous
       paragraph. The trace contains entries for function calls,
       returns to function, process scheduling, other process related
       (spawn, etc) events, and garbage collection. All trace entries
       are timestamped.</item>
      <tag><c>2</c></tag>
      <item>Profiling; the trace file is read, the execution call
       stack is simulated, and raw profile data is calculated from
       the simulated call stack and the trace timestamps. The profile
       data is stored in the <c>fprof</c> server state. During this
       step the trace data may be dumped in text format to file or
       console. </item>
      <tag><c>3</c></tag>
      <item>Analysing; the raw profile data is sorted, filtered and
       dumped in text format either to file or console. The text
       format intended to be both readable for a human reader, as
       well as parsable with the standard erlang parsing tools.</item>
    </taglist>
    <p>Since <c>fprof</c> uses trace to file, the runtime performance
      degradation is minimized, but still far from negligible,
      especially for programs that use the filesystem heavily by
      themselves. Where you place the trace file is also important,
      e.g on Solaris <c>/tmp</c> is usually a good choice since it is
      essentially a RAM disk, while any NFS (network) mounted disk is
      a bad idea.
      </p>
    <p><c>fprof</c> can also skip the file step and trace to a tracer
      process that does the profiling in runtime.
            <marker id="start"></marker>
</p>
  </description>
  <funcs>
    <func>
      <name>start() -> {ok, Pid} | {error, {already_started, Pid}}</name>
      <fsummary>Starts the <c>fprof</c>&nbsp;server.</fsummary>
      <type>
        <v>Pid = pid()</v>
      </type>
      <desc>
        <p>Starts the <c>fprof</c>&nbsp;server. 
          </p>
        <p>Note that it seldom
          needs to be started explicitly since it is automatically
          started by the functions that need a running server.
                    <marker id="stop"></marker>
</p>
      </desc>
    </func>
    <func>
      <name>stop() -> ok</name>
      <fsummary>Same as <c>stop(normal)</c>.</fsummary>
      <desc>
        <p>Same as <c>stop(normal)</c>.</p>
      </desc>
    </func>
    <func>
      <name>stop(Reason) -> ok</name>
      <fsummary>Stops the <c>fprof</c>&nbsp;server.</fsummary>
      <type>
        <v>Reason = term()</v>
      </type>
      <desc>
        <p>Stops the <c>fprof</c>&nbsp;server.
          </p>
        <p>The supplied <c>Reason</c> becomes the exit reason for the
          server process. Default Any
          <c>Reason</c> other than <c>kill</c> sends a request to the
          server and waits for it to clean up, reply and exit. If
          <c>Reason</c> is <c>kill</c>, the server is bluntly killed.
          </p>
        <p>If the <c>fprof</c>&nbsp;server is not running, this
          function returns immediately with the same return value.
          </p>
        <note>
          <p>When the <c>fprof</c>&nbsp;server is stopped the
            collected raw profile data is lost.</p>
        </note>
        <marker id="apply"></marker>
      </desc>
    </func>
    <func>
      <name>apply(Func, Args) -> term()</name>
      <fsummary>Same as <c>apply(Func, Args, [])</c>.</fsummary>
      <type>
        <v>Func = function() | {Module, Function}</v>
        <v>Args = [term()]</v>
        <v>Module = atom()</v>
        <v>Function = atom()</v>
      </type>
      <desc>
        <p>Same as <c>apply(Func, Args, [])</c>.</p>
      </desc>
    </func>
    <func>
      <name>apply(Module, Function, Args) -> term()</name>
      <fsummary>Same as <c>apply({Module, Function}, Args, [])</c>.</fsummary>
      <type>
        <v>Args = [term()]</v>
        <v>Module = atom()</v>
        <v>Function = atom()</v>
      </type>
      <desc>
        <p>Same as <c>apply({Module, Function}, Args, [])</c>.</p>
      </desc>
    </func>
    <func>
      <name>apply(Func, Args, OptionList) -> term()</name>
      <fsummary>Calls <c>erlang:apply(Func, Args)</c>surrounded by<c>trace([start | OptionList])</c>and<c>trace(stop)</c>.</fsummary>
      <type>
        <v>Func = function() | {Module, Function}</v>
        <v>Args = [term()]</v>
        <v>OptionList = [Option]</v>
        <v>Module = atom()</v>
        <v>Function = atom()</v>
        <v>Option = continue | start | {procs, PidList} | TraceStartOption</v>
      </type>
      <desc>
        <p>Calls <c>erlang:apply(Func, Args)</c> surrounded by
          <c>trace([start, ...])</c> and
          <c>trace(stop)</c>.
          </p>
        <p>Some effort is made to keep the trace clean from unnecessary
          trace messages; tracing is started and stopped from a spawned
          process while the <c>erlang:apply/2</c> call is made in the
          current process, only surrounded by <c>receive</c> and
          <c>send</c> statements towards the trace starting
          process. The trace starting process exits when not needed
          any more.
          </p>
        <p>The <c>TraceStartOption</c> is any option allowed for 
          <c>trace/1</c>. The options 
          <c>[start, {procs, [self() | PidList]} | OptList]</c> 
          are given to <c>trace/1</c>, where <c>OptList</c> is 
          <c>OptionList</c> with <c>continue</c>, <c>start</c> 
          and <c>{procs, _}</c> options removed.
          </p>
        <p>The <c>continue</c> option inhibits the call to
          <c>trace(stop)</c> and leaves it up to the caller to stop
          tracing at a suitable time.</p>
      </desc>
    </func>
    <func>
      <name>apply(Module, Function, Args, OptionList) -> term()</name>
      <fsummary>Same as <c>apply({Module, Function}, Args, OptionList)</c>.</fsummary>
      <type>
        <v>Module = atom()</v>
        <v>Function = atom()</v>
        <v>Args = [term()]</v>
      </type>
      <desc>
        <p>Same as 
          <c>apply({Module, Function}, Args, OptionList)</c>.
          </p>
        <p><c>OptionList</c> is an option list allowed for 
          <c>apply/3</c>.
                    <marker id="trace"></marker>
</p>
      </desc>
    </func>
    <func>
      <name>trace(start, Filename) -> ok | {error, Reason} |  {'EXIT', ServerPid, Reason}</name>
      <fsummary>Same as <c>trace([start, {file, Filename}])</c>.</fsummary>
      <type>
        <v>Reason = term()</v>
      </type>
      <desc>
        <p>Same as <c>trace([start, {file, Filename}])</c>.</p>
      </desc>
    </func>
    <func>
      <name>trace(verbose, Filename) -> ok | {error, Reason} |  {'EXIT', ServerPid, Reason}</name>
      <fsummary>Same as <c>trace([start, verbose, {file, Filename}])</c>.</fsummary>
      <type>
        <v>Reason = term()</v>
      </type>
      <desc>
        <p>Same as 
          <c>trace([start, verbose, {file, Filename}])</c>.</p>
      </desc>
    </func>
    <func>
      <name>trace(OptionName, OptionValue) -> ok | {error, Reason} |  {'EXIT', ServerPid, Reason}</name>
      <fsummary>Same as <c>trace([{OptionName, OptionValue}])</c>.</fsummary>
      <type>
        <v>OptionName = atom()</v>
        <v>OptionValue = term()</v>
        <v>Reason = term()</v>
      </type>
      <desc>
        <p>Same as 
          <c>trace([{OptionName, OptionValue}])</c>.</p>
      </desc>
    </func>
    <func>
      <name>trace(verbose) -> ok | {error, Reason} |  {'EXIT', ServerPid, Reason}</name>
      <fsummary>Same as <c>trace([start, verbose])</c>.</fsummary>
      <type>
        <v>Reason = term()</v>
      </type>
      <desc>
        <p>Same as <c>trace([start, verbose])</c>.</p>
      </desc>
    </func>
    <func>
      <name>trace(OptionName) -> ok | {error, Reason} |  {'EXIT', ServerPid, Reason}</name>
      <fsummary>Same as <c>trace([OptionName])</c>.</fsummary>
      <type>
        <v>OptionName = atom()</v>
        <v>Reason = term()</v>
      </type>
      <desc>
        <p>Same as <c>trace([OptionName])</c>.</p>
      </desc>
    </func>
    <func>
      <name>trace({OptionName, OptionValue}) -> ok | {error, Reason} |  {'EXIT', ServerPid, Reason}</name>
      <fsummary>Same as <c>trace([{OptionName, OptionValue}])</c>.</fsummary>
      <type>
        <v>OptionName = atom()</v>
        <v>OptionValue = term()</v>
        <v>Reason = term()</v>
      </type>
      <desc>
        <p>Same as 
          <c>trace([{OptionName, OptionValue}])</c>.</p>
      </desc>
    </func>
    <func>
      <name>trace([Option]) -> ok | {error, Reason} |  {'EXIT', ServerPid, Reason}</name>
      <fsummary>Starts or stops tracing.</fsummary>
      <type>
        <v>Option = start | stop | {procs, PidSpec} | {procs, [PidSpec]} | verbose | {verbose, bool()} |  file | {file, Filename} | {tracer, Tracer}</v>
        <v>PidSpec = pid() | atom()</v>
        <v>Tracer = pid() | port()</v>
        <v>Reason = term()</v>
      </type>
      <desc>
        <p>Starts or stops tracing.
          </p>
        <p><c>PidSpec</c> and <c>Tracer</c> are used in calls to
          <c>erlang:trace(PidSpec, true, [{tracer, Tracer} | Flags])</c>, and <c>Filename</c> is used to call 
          <c>dbg:trace_port(file, Filename)</c>. Please see the
          appropriate documentation.</p>
        <p>Option description:</p>
        <taglist>
          <tag><c>stop</c></tag>
          <item>Stops a running <c>fprof</c> trace and clears all tracing
           from the node. Either option <c>stop</c> or <c>start</c> must be
           specified, but not both.</item>
          <tag><c>start</c></tag>
          <item>Clears all tracing from the node and starts a new
          <c>fprof</c> trace. Either option <c>start</c> or
          <c>stop</c> must be specified, but not both.</item>
          <tag><c>verbose</c>| <c>{verbose, bool()}</c></tag>
          <item>The options <c>verbose</c> or <c>{verbose, true}</c>
           adds some trace flags that <c>fprof</c> does not need, but
           that may be interesting for general debugging
           purposes. This option is only
           allowed with the <c>start</c> option.</item>
          <tag><c>cpu_time</c>| <c>{cpu_time, bool()}</c></tag>
          <item>The options <c>cpu_time</c> or <c>{cpu_time, true></c>
           makes the timestamps in the trace be in CPU time instead
           of wallclock time which is the default. This option is
           only allowed with the <c>start</c> option.</item>
          <tag><c>{procs, PidSpec}</c>| <c>{procs, [PidSpec]}</c></tag>
          <item>Specifies which processes that shall be traced. If
           this option is not given, the calling process is
           traced. All processes spawned by the traced processes are
           also traced.
           This option is only allowed with the <c>start</c> option.</item>
          <tag><c>file</c>| <c>{file, Filename}</c></tag>
          <item>Specifies the filename of the trace. 
           If the option <c>file</c> is given, or none of these
           options are given, the file <c>"fprof.trace"</c> is used.
           This option is only allowed with the <c>start</c> option,
           but not with the <c>{tracer, Tracer}</c> option.</item>
          <tag><c>{tracer, Tracer}</c></tag>
          <item>Specifies that trace to process or port shall be done
           instead of trace to file.
           This option is only allowed with the <c>start</c> option,
           but not with the <c>{file, Filename}</c> option.</item>
        </taglist>
        <marker id="profile"></marker>
      </desc>
    </func>
    <func>
      <name>profile() -> ok | {error, Reason} |  {'EXIT', ServerPid, Reason}</name>
      <fsummary>Same as <c>profile([])</c>.</fsummary>
      <type>
        <v>Reason = term()</v>
      </type>
      <desc>
        <p>Same as <c>profile([])</c>.</p>
      </desc>
    </func>
    <func>
      <name>profile(OptionName, OptionValue) -> ok | {error, Reason} |  {'EXIT', ServerPid, Reason}</name>
      <fsummary>Same as <c>profile([{OptionName, OptionValue}])</c>.</fsummary>
      <type>
        <v>OptionName = atom()</v>
        <v>OptionValue = term()</v>
        <v>Reason = term()</v>
      </type>
      <desc>
        <p>Same as 
          <c>profile([{OptionName, OptionValue}])</c>.</p>
      </desc>
    </func>
    <func>
      <name>profile(OptionName) -> ok | {error, Reason} |  {'EXIT', ServerPid, Reason}</name>
      <fsummary>Same as <c>profile([OptionName])</c>.</fsummary>
      <type>
        <v>OptionName = atom()</v>
        <v>Reason = term()</v>
      </type>
      <desc>
        <p>Same as <c>profile([OptionName])</c>.</p>
      </desc>
    </func>
    <func>
      <name>profile({OptionName, OptionValue}) -> ok | {error, Reason} |  {'EXIT', ServerPid, Reason}</name>
      <fsummary>Same as <c>profile([{OptionName, OptionValue}])</c>.</fsummary>
      <type>
        <v>OptionName = atom()</v>
        <v>OptionValue = term()</v>
        <v>Reason = term()</v>
      </type>
      <desc>
        <p>Same as 
          <c>profile([{OptionName, OptionValue}])</c>.</p>
      </desc>
    </func>
    <func>
      <name>profile([Option]) -> ok | {ok, Tracer} | {error, Reason} |  {'EXIT', ServerPid, Reason}</name>
      <fsummary>Compiles a trace into raw profile data held by the <c>fprof</c>&nbsp;server.</fsummary>
      <type>
        <v>Option = file | {file, Filename} | dump | {dump, Dump} |  append | start | stop</v>
        <v>Dump = pid() | Dumpfile | []</v>
        <v>Tracer = pid()</v>
        <v>Reason = term()</v>
      </type>
      <desc>
        <p>Compiles a trace into raw profile data held by the 
          <c>fprof</c>&nbsp;server.
          </p>
        <p><c>Dumpfile</c> is used to call <c>file:open/2</c>, 
          and <c>Filename</c> is used to call 
          <c>dbg:trace_port(file, Filename)</c>. Please see the
          appropriate documentation.</p>
        <p>Option description:</p>
        <taglist>
          <tag><c>file</c>| <c>{file, Filename}</c></tag>
          <item>Reads the file <c>Filename</c> and creates raw
           profile data that is stored in RAM by the
          <c>fprof</c>&nbsp;server. If the option <c>file</c> is
           given, or none of these options are given, the file
          <c>"fprof.trace"</c> is read. The call will return when
           the whole trace has been
           read with the return value <c>ok</c> if successful.
           This option is not allowed with the <c>start</c> or
          <c>stop</c> options.</item>
          <tag><c>dump</c>| <c>{dump, Dump}</c></tag>
          <item>Specifies the destination for the trace text dump. If
           this option is not given, no dump is generated, if it is
          <c>dump</c> the destination will be the
           caller's group leader, otherwise the destination
          <c>Dump</c> is either the pid of an I/O device or
           a filename. And, finally, if the filename is <c>[]</c> -
          <c>"fprof.dump"</c> is used instead.
           This option is not allowed with the <c>stop</c> option.</item>
          <tag><c>append</c></tag>
          <item>Causes the trace text dump to be appended to the
           destination file.
           This option is only allowed with the  
          <c>{dump, Dumpfile}</c> option.</item>
          <tag><c>start</c></tag>
          <item>Starts a tracer process that profiles trace data in
           runtime. The call will return immediately with the return
           value <c>{ok, Tracer}</c> if successful.
           This option is not allowed with the <c>stop</c>,
          <c>file</c> or  <c>{file, Filename}</c> options.</item>
          <tag><c>stop</c></tag>
          <item>Stops the tracer process that profiles trace data in
           runtime. The return value will be value <c>ok</c> if successful.
           This option is not allowed with the <c>start</c>,
          <c>file</c> or <c>{file, Filename}</c> options.</item>
        </taglist>
        <marker id="analyse"></marker>
      </desc>
    </func>
    <func>
      <name>analyse() -> ok | {error, Reason} |  {'EXIT', ServerPid, Reason}</name>
      <fsummary>Same as <c>analyse([])</c>.</fsummary>
      <type>
        <v>Reason = term()</v>
      </type>
      <desc>
        <p>Same as <c>analyse([])</c>.</p>
      </desc>
    </func>
    <func>
      <name>analyse(OptionName, OptionValue) -> ok | {error, Reason} |  {'EXIT', ServerPid, Reason}</name>
      <fsummary>Same as <c>analyse([{OptionName, OptionValue}])</c>.</fsummary>
      <type>
        <v>OptionName = atom()</v>
        <v>OptionValue = term()</v>
        <v>Reason = term()</v>
      </type>
      <desc>
        <p>Same as 
          <c>analyse([{OptionName, OptionValue}])</c>.</p>
      </desc>
    </func>
    <func>
      <name>analyse(OptionName) -> ok | {error, Reason} |  {'EXIT', ServerPid, Reason}</name>
      <fsummary>Same as <c>analyse([OptionName])</c>.</fsummary>
      <type>
        <v>OptionName = atom()</v>
        <v>Reason = term()</v>
      </type>
      <desc>
        <p>Same as <c>analyse([OptionName])</c>.</p>
      </desc>
    </func>
    <func>
      <name>analyse({OptionName, OptionValue}) -> ok | {error, Reason} |  {'EXIT', ServerPid, Reason}</name>
      <fsummary>Same as <c>analyse([{OptionName, OptionValue}])</c>.</fsummary>
      <type>
        <v>OptionName = atom()</v>
        <v>OptionValue = term()</v>
        <v>Reason = term()</v>
      </type>
      <desc>
        <p>Same as 
          <c>analyse([{OptionName, OptionValue}])</c>.</p>
      </desc>
    </func>
    <func>
      <name>analyse([Option]) -> ok | {error, Reason} |  {'EXIT', ServerPid, Reason}</name>
      <fsummary>Analyses raw profile data in the <c>fprof</c>&nbsp;server.</fsummary>
      <type>
        <v>Option = dest | {dest, Dest} | append | {cols, Cols} |  callers | {callers, bool()} | no_callers | {sort, SortSpec} | totals | {totals, bool()} |  details | {details, bool()} | no_details</v>
        <v>Dest = pid() | Destfile</v>
        <v>Cols = integer() >= 80</v>
        <v>SortSpec = acc | own</v>
        <v>Reason = term()</v>
      </type>
      <desc>
        <p>Analyses raw profile data in the
          <c>fprof</c>&nbsp;server. If called while there is no raw
          profile data available, <c>{error, no_profile}</c> is
          returned. 
          </p>
        <p><c>Destfile</c> is used to call <c>file:open/2</c>. 
          Please see the appropriate documentation.</p>
        <p>Option description:</p>
        <taglist>
          <tag><c>dest</c>| <c>{dest, Dest}</c></tag>
          <item>Specifies the destination for the analysis. If
           this option is not given or it is <c>dest</c>, 
           the destination will be the caller's group leader, 
           otherwise the destination <c>Dest</c> is either 
           the <c>pid()</c> of an I/O device or a filename. 
           And, finally, if the filename is <c>[]</c> -
          <c>"fprof.analysis"</c> is used instead.</item>
          <tag><c>append</c></tag>
          <item>Causes the analysis to be appended to the
           destination file.
           This option is only allowed with the  
          <c>{dest, Destfile}</c> option.</item>
          <tag><c>{cols, Cols}</c></tag>
          <item>Specifies the number of columns in the analysis text.
           If this option is not given the number of columns is set
           to 80.</item>
          <tag><c>callers</c>| <c>{callers, true}</c></tag>
          <item>Prints callers and called information in the
           analysis. This is the default.</item>
          <tag><c>{callers, false}</c>| <c>no_callers</c></tag>
          <item>Suppresses the printing of callers and called
           information in the analysis.</item>
          <tag><c>{sort, SortSpec}</c></tag>
          <item>Specifies if the analysis should be sorted according
           to the ACC column, which is the default, or the OWN
           column. See 
          <seealso marker="#analysis">Analysis Format</seealso> below.</item>
          <tag><c>totals</c>| <c>{totals, true}</c></tag>
          <item>Includes a section containing call statistics
           for all calls regardless of process, in the analysis.</item>
          <tag><c>{totals, false}</c></tag>
          <item>Supresses the totals section in the analysis, which is
           the default.</item>
          <tag><c>details</c>| <c>{details, true}</c></tag>
          <item>Prints call statistics for each process in the
           analysis. This is the default.</item>
          <tag><c>{details, false}</c>| <c>no_details</c></tag>
          <item>Suppresses the call statistics for each process from
           the analysis.</item>
        </taglist>
      </desc>
    </func>
  </funcs>

  <section>
    <marker id="analysis"></marker>
    <title>Analysis format</title>
    <p>This section describes the output format of the analyse
      command. See <seealso marker="#analyse">analyse/0</seealso>.
      </p>
    <p>The format is parsable with the standard Erlang parsing tools
      <c>erl_scan</c> and <c>erl_parse</c>, <c>file:consult/1</c> or
      <c>io:read/2</c>. The parse format is not explained here - it
      should be easy for the interested to try it out. Note that some
      flags to <c>analyse/1</c> will affect the format. 
      </p>
    <p>The following example was run on OTP/R8 on Solaris 8, all OTP
      internals in this example are very version dependent.
      </p>
    <p>As an example, we will use the following function, that you may
      recognise as a slightly modified benchmark function from the
      manpage file(3):</p>
    <code type="none"><![CDATA[
-module(foo).
-export([create_file_slow/2]).

create_file_slow(Name, N) when integer(N), N >= 0 ->
    {ok, FD} = 
        file:open(Name, [raw, write, delayed_write, binary]),
    if N > 256 ->
            ok = file:write(FD, 
                            lists:map(fun (X) -> <<X:32/unsigned>> end,
                            lists:seq(0, 255))),
            ok = create_file_slow(FD, 256, N);
       true ->
            ok = create_file_slow(FD, 0, N)
    end,
    ok = file:close(FD).

create_file_slow(FD, M, M) ->
    ok;
create_file_slow(FD, M, N) ->
    ok = file:write(FD, <<M:32/unsigned>>),
    create_file_slow(FD, M+1, N).]]></code>
    <p>Let us have a look at the printout after running:</p>
    <pre>
1> <input>fprof:apply(foo, create_file_slow, [junk, 1024]).</input>
2> <input>fprof:profile().</input>
3> <input>fprof:analyse().</input></pre>
    <p>The printout starts with:</p>
    <pre>
%% Analysis results:
{  analysis_options,
 [{callers, true},
  {sort, acc},
  {totals, false},
  {details, true}]}.

%                                       CNT       ACC       OWN        
[{ totals,                             9627, 1691.119, 1659.074}].  %%%</pre>
    <p>The CNT column shows the total number of function calls that
      was found in the trace. In the ACC column is the total time of
      the trace from first timestamp to last. And in the OWN
      column is the sum of the execution time in functions found in the
      trace, not including called functions. In this case it is very
      close to the ACC time since the emulator had practically nothing
      else to do than to execute our test program.
      </p>
    <p>All time values in the printout are in milliseconds.
      </p>
    <p>The printout continues:</p>
    <pre>
%                                       CNT       ACC       OWN        
[{ "&lt;0.28.0>",                         9627,undefined, 1659.074}].   %%</pre>
    <p>This is the printout header of one process. The printout
      contains only this one process since we did <c>fprof:apply/3</c>
      which traces only the current process. Therefore the CNT and
      OWN columns perfectly matches the totals above. The ACC column is
      undefined since summing the ACC times of all calls in the process
      makes no sense - you would get something like the ACC value from
      totals above multiplied by the average depth of the call stack,
      or something.
      </p>
    <p>All paragraphs up to the next process header only concerns
      function calls within this process.
      </p>
    <p>Now we come to something more interesting:</p>
    <pre>
{[{undefined,                             0, 1691.076,    0.030}],     
 { {fprof,apply_start_stop,4},            0, 1691.076,    0.030},     %
 [{{foo,create_file_slow,2},              1, 1691.046,    0.103},      
  {suspend,                               1,    0.000,    0.000}]}.    

{[{{fprof,apply_start_stop,4},            1, 1691.046,    0.103}],     
 { {foo,create_file_slow,2},              1, 1691.046,    0.103},     %
 [{{file,close,1},                        1, 1398.873,    0.019},      
  {{foo,create_file_slow,3},              1,  249.678,    0.029},      
  {{file,open,2},                         1,   20.778,    0.055},      
  {{lists,map,2},                         1,   16.590,    0.043},      
  {{lists,seq,2},                         1,    4.708,    0.017},      
  {{file,write,2},                        1,    0.316,    0.021}]}.    </pre>
    <p>The printout consists of one paragraph per called function. The
      function <em>marked</em> with '%' is the one the paragraph
      concerns - <c>foo:create_file_slow/2</c>. Above the marked
      function are the <em>calling</em> functions -  those that has
      called the marked, and below are those <em>called</em> by the
      marked function. 
      </p>
    <p>The paragraphs are per default sorted in decreasing order of
      the ACC column for the marked function. The calling list and
      called list within one paragraph are also per default sorted in
      decreasing order of their ACC column.
      </p>
    <p>The columns are: CNT - the number of times the function
      has been called, ACC - the time spent in the
      function including called functions, and OWN - the
      time spent in the function not including called
      functions. 
      </p>
    <p>The rows for the <em>calling</em> functions contain statistics
      for the <em>marked</em> function with the constraint that only
      the occasions when a call was made from the <em>row's</em>
      function to the <em>marked</em> function are accounted for.
      </p>
    <p>The row for the <em>marked</em> function simply contains the
      sum of all <em>calling</em> rows.  
      </p>
    <p>The rows for the <em>called</em> functions contains statistics
      for the <em>row's</em> function with the constraint that only the
      occasions when a call was made from the <em>marked</em> to the
      <em>row's</em> function are accounted for.
      </p>
    <p>So, we see that <c>foo:create_file_slow/2</c> used very little
      time for its own execution. It spent most of its time in
      <c>file:close/1</c>. The function <c>foo:create_file_slow/3</c>
      that writes 3/4 of the file contents is the second biggest time
      thief. 
      </p>
    <p>We also see that the call to <c>file:write/2</c> that writes
      1/4 of the file contents takes very little time in itself. What
      takes time is to build the data (<c>lists:seq/2</c> and
      <c>lists:map/2</c>).
      </p>
    <p>The function 'undefined' that has called
      <c>fprof:apply_start_stop/4</c> is an unknown function because that
      call was not recorded in the trace. It was only recorded
      that the execution returned from
      <c>fprof:apply_start_stop/4</c> to some other function above in
      the call stack, or that the process exited from there.
      </p>
    <p>Let us continue down the printout to find:</p>
    <pre>
{[{{foo,create_file_slow,2},              1,  249.678,    0.029},      
  {{foo,create_file_slow,3},            768,    0.000,   23.294}],     
 { {foo,create_file_slow,3},            769,  249.678,   23.323},     %
 [{{file,write,2},                      768,  220.314,   14.539},      
  {suspend,                              57,    6.041,    0.000},      
  {{foo,create_file_slow,3},            768,    0.000,   23.294}]}.    </pre>
    <p>If you compare with the code you will see there also that
      <c>foo:create_file_slow/3</c> was called only from
      <c>foo:create_file_slow/2</c> and itself, and called only
      <c>file:write/2</c>, note the number of calls to
      <c>file:write/2</c>. But here we see that <c>suspend</c> was
      called a few times. This is a pseudo function that indicates
      that the process was suspended while executing in
      <c>foo:create_file_slow/3</c>, and since there is no
      <c>receive</c> or <c>erlang:yield/0</c> in the code, it must be
      Erlang scheduling suspensions, or the trace file driver
      compensating for large file write operations (these are regarded
      as a schedule out followed by a schedule in to the same process).
      </p>
    <p></p>
    <p>Let us find the <c>suspend</c> entry:</p>
    <pre>
{[{{file,write,2},                       53,    6.281,    0.000},      
  {{foo,create_file_slow,3},             57,    6.041,    0.000},      
  {{prim_file,drv_command,4},            50,    4.582,    0.000},      
  {{prim_file,drv_get_response,1},       34,    2.986,    0.000},      
  {{lists,map,2},                        10,    2.104,    0.000},      
  {{prim_file,write,2},                  17,    1.852,    0.000},      
  {{erlang,port_command,2},              15,    1.713,    0.000},      
  {{prim_file,drv_command,2},            22,    1.482,    0.000},      
  {{prim_file,translate_response,2},     11,    1.441,    0.000},      
  {{prim_file,'-drv_command/2-fun-0-',1},  15,    1.340,    0.000},      
  {{lists,seq,4},                         3,    0.880,    0.000},      
  {{foo,'-create_file_slow/2-fun-0-',1},   5,    0.523,    0.000},      
  {{erlang,bump_reductions,1},            4,    0.503,    0.000},      
  {{prim_file,open_int_setopts,3},        1,    0.165,    0.000},      
  {{prim_file,i32,4},                     1,    0.109,    0.000},      
  {{fprof,apply_start_stop,4},            1,    0.000,    0.000}],     
 { suspend,                             299,   32.002,    0.000},     %
 [ ]}.</pre>
    <p>We find no particulary long suspend times, so no function seems
      to have waited in a receive statement. Actually,
      <c>prim_file:drv_command/4</c> contains a receive statement, but
      in this test program, the message lies in the process receive
      buffer when the receive statement is entered. We also see that
      the total suspend time for the test run is small.
      </p>
    <p>The <c>suspend</c> pseudo function has got an OWN time of
      zero. This is to prevent the process total OWN time from
      including time in suspension. Whether suspend time is really ACC
      or OWN time is more of a philosophical question.
      </p>
    <p>Now we look at another interesting pseudo function,
      <c>garbage_collect</c>:</p>
    <pre>
{[{{prim_file,drv_command,4},            25,    0.873,    0.873},      
  {{prim_file,write,2},                  16,    0.692,    0.692},      
  {{lists,map,2},                         2,    0.195,    0.195}],     
 { garbage_collect,                      43,    1.760,    1.760},     %
 [ ]}.</pre>
    <p>Here we see that no function distinguishes itself considerably,
      which is very normal.
      </p>
    <p>The <c>garbage_collect</c> pseudo function has not got an OWN
      time of zero like <c>suspend</c>, instead it is equal to the ACC
      time. 
      </p>
    <p>Garbage collect often occurs while a process is suspended, but
      <c>fprof</c> hides this fact by pretending that the suspended
      function was first unsuspended and then garbage
      collected. Otherwise the printout would show
      <c>garbage_collect</c> being called from <c>suspend</c> but not
      not which function that might have caused the garbage
      collection. 
      </p>
    <p>Let us now get back to the test code:</p>
    <pre>
{[{{foo,create_file_slow,3},            768,  220.314,   14.539},      
  {{foo,create_file_slow,2},              1,    0.316,    0.021}],     
 { {file,write,2},                      769,  220.630,   14.560},     %
 [{{prim_file,write,2},                 769,  199.789,   22.573},      
  {suspend,                              53,    6.281,    0.000}]}.    </pre>
    <p>Not unexpectedly, we see that <c>file:write/2</c> was called
      from <c>foo:create_file_slow/3</c> and
      <c>foo:create_file_slow/2</c>. The number of calls in each case as
      well as the used time are also just confirms the previous results.
      </p>
    <p>We see that <c>file:write/2</c> only calls
      <c>prim_file:write/2</c>, but let us refrain from digging into the
      internals of the kernel application.
      </p>
    <p>But, if we nevertheless <em>do</em> dig down we find
      the call to the linked in driver that does the file operations
      towards the host operating system:</p>
    <pre>
{[{{prim_file,drv_command,4},           772, 1458.356, 1456.643}],     
 { {erlang,port_command,2},             772, 1458.356, 1456.643},     %
 [{suspend,                              15,    1.713,    0.000}]}.    </pre>
    <p>This is 86 % of the total run time, and as we saw before it
      is the close operation the absolutely biggest contributor. We
      find a comparison ratio a little bit up in the call stack:</p>
    <pre>
{[{{prim_file,close,1},                   1, 1398.748,    0.024},      
  {{prim_file,write,2},                 769,  174.672,   12.810},      
  {{prim_file,open_int,4},                1,   19.755,    0.017},      
  {{prim_file,open_int_setopts,3},        1,    0.147,    0.016}],     
 { {prim_file,drv_command,2},           772, 1593.322,   12.867},     %
 [{{prim_file,drv_command,4},           772, 1578.973,   27.265},      
  {suspend,                              22,    1.482,    0.000}]}.    </pre>
    <p>The time for file operations in the linked in driver
      distributes itself as 1 % for open, 11 % for write and 87 % for
      close. All data is probably buffered in the operating system
      until the close.
      </p>
    <p>The unsleeping reader may notice that the ACC times for 
      <c>prim_file:drv_command/2</c> and
      <c>prim_file:drv_command/4</c> is not equal between the
      paragraphs above, even though it is easy to believe that
      <c>prim_file:drv_command/2</c> is just a passthrough function.
      </p>
    <p>The missing time can be found in the paragraph
      for <c>prim_file:drv_command/4</c> where it is evident that not
      only <c>prim_file:drv_command/2</c> is called but also a fun:
      </p>
    <pre>
{[{{prim_file,drv_command,2},           772, 1578.973,   27.265}],     
 { {prim_file,drv_command,4},           772, 1578.973,   27.265},     %
 [{{erlang,port_command,2},             772, 1458.356, 1456.643},      
  {{prim_file,'-drv_command/2-fun-0-',1}, 772,   87.897,   12.736},      
  {suspend,                              50,    4.582,    0.000},      
  {garbage_collect,                      25,    0.873,    0.873}]}.    </pre>
    <p>And some more missing time can be explained by the fact that
      <c>prim_file:open_int/4</c> both calls
      <c>prim_file:drv_command/2</c> directly as well as through
      <c>prim_file:open_int_setopts/3</c>, which complicates the
      picture. 
      </p>
    <pre>
{[{{prim_file,open,2},                    1,   20.309,    0.029},      
  {{prim_file,open_int,4},                1,    0.000,    0.057}],     
 { {prim_file,open_int,4},                2,   20.309,    0.086},     %
 [{{prim_file,drv_command,2},             1,   19.755,    0.017},      
  {{prim_file,open_int_setopts,3},        1,    0.360,    0.032},      
  {{prim_file,drv_open,2},                1,    0.071,    0.030},      
  {{erlang,list_to_binary,1},             1,    0.020,    0.020},      
  {{prim_file,i32,1},                     1,    0.017,    0.017},      
  {{prim_file,open_int,4},                1,    0.000,    0.057}]}.    
.
.
.
{[{{prim_file,open_int,4},                1,    0.360,    0.032},      
  {{prim_file,open_int_setopts,3},        1,    0.000,    0.016}],     
 { {prim_file,open_int_setopts,3},        2,    0.360,    0.048},     %
 [{suspend,                               1,    0.165,    0.000},      
  {{prim_file,drv_command,2},             1,    0.147,    0.016},      
  {{prim_file,open_int_setopts,3},        1,    0.000,    0.016}]}.    </pre>
  </section>

  <section>
    <title>Notes</title>
    <p>The actual supervision of execution times is in itself a
      CPU intensive activity. A message is written on the trace file
      for every function call that is made by the profiled code.
      </p>
    <p>The ACC time calculation is sometimes difficult to make
      correct, since it is difficult to define. This happens
      especially when a function occurs in several instances in the
      call stack, for example by calling itself perhaps through other
      functions and perhaps even non-tail recursively.
      </p>
    <p>To produce sensible results, <c>fprof</c> tries not to charge
      any function more than once for ACC time. The instance highest
      up (with longest duration) in the call stack is chosen.
      </p>
    <p>Sometimes a function may unexpectedly waste a lot (some 10 ms
      or more depending on host machine OS) of OWN (and ACC) time, even
      functions that does practically nothing at all. The problem may
      be that the OS has chosen to schedule out the
      Erlang runtime system process for a while, and if the OS does
      not support high resolution cpu time measurements
      <c>fprof</c> will use wallclock time for its calculations, and
      it will appear as functions randomly burn virtual machine time.</p>
  </section>

  <section>
    <title>See Also</title>
    <p>dbg(3), <seealso marker="eprof">eprof</seealso>(3), erlang(3), 
      io(3), 
      <seealso marker="fprof_chapter">Tools User's Guide</seealso></p>
  </section>
</erlref>