<?xml version="1.0" encoding="utf-8" ?>
<!DOCTYPE erlref SYSTEM "erlref.dtd">
<erlref>
<header>
<copyright>
<year>2001</year><year>2013</year>
<holder>Ericsson AB. All Rights Reserved.</holder>
</copyright>
<legalnotice>
Licensed under the Apache License, Version 2.0 (the "License");
you may not use this file except in compliance with the License.
You may obtain a copy of the License at
http://www.apache.org/licenses/LICENSE-2.0
Unless required by applicable law or agreed to in writing, software
distributed under the License is distributed on an "AS IS" BASIS,
WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
See the License for the specific language governing permissions 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> server.</fsummary>
<type>
<v>Pid = pid()</v>
</type>
<desc>
<p>Starts the <c>fprof</c> 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> server.</fsummary>
<type>
<v>Reason = term()</v>
</type>
<desc>
<p>Stops the <c>fprof</c> 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> server is not running, this
function returns immediately with the same return value.
</p>
<note>
<p>When the <c>fprof</c> 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> 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> 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> 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> 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> 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
[{ "<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
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>
