path: root/lib/os_mon/doc/src/cpu_sup.xml

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

<erlref>
  <header>
    <copyright>
      <year>1997</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>cpu_sup</title>
    <prepared></prepared>
    <docno></docno>
    <date></date>
    <rev></rev>
  </header>
  <module>cpu_sup</module>
  <modulesummary>A CPU Load and CPU Utilization Supervisor Process</modulesummary>
  <description>
    <p><c>cpu_sup</c> is a process which supervises the CPU load
      and CPU utilization. It is part of the OS_Mon application, see
      <seealso marker="os_mon_app">os_mon(6)</seealso>. Available for Unix,
      although CPU utilization values (<c>util/0,1</c>) are only
      available for Solaris and Linux.</p>
    <p>The load values are proportional to how long time a runnable
      Unix process has to spend in the run queue before it is scheduled.
      Accordingly, higher values mean more system load. The returned
      value divided by 256 produces the figure displayed by <c>rup</c>
      and <c>top</c>. What is displayed as 2.00 in <c>rup</c>, is
      displayed as load up to the second mark in <c>xload</c>.</p>
    <p>For example, <c>rup</c> displays a load of 128 as 0.50, and
      512 as 2.00.</p>
    <p>If the user wants to view load values as percentage of machine
      capacity, then this way of measuring presents a problem, because
      the load values are not restricted to a fixed interval. In this
      case, the following simple mathematical transformation can
      produce the load value as a percentage:</p>
    <code type="none">
      PercentLoad = 100 * (1 - D/(D + Load))
    </code>
    <p><c>D</c> determines which load value should be associated with
      which percentage. Choosing <c>D</c> = 50 means that 128 is 60% 
      load, 256 is 80%, 512 is 90%, and so on.</p>
    <p>Another way of measuring system load is to divide the number of
      busy CPU cycles by the total number of CPU cycles. This produces 
      values in the 0-100 range immediately. However, this method hides 
      the fact that a machine can be more or less saturated. CPU
      utilization is therefore a better name than system load for this
      measure.</p>
    <p>A server which receives just enough requests to never become
      idle will score a CPU utilization of 100%. If the server receives
      50% more requests, it will still scores 100%. When the system load
      is calculated with the percentage formula shown previously,
      the load will increase from 80% to 87%.</p>
    <p>The <c>avg1/0</c>, <c>avg5/0</c>, and <c>avg15/0</c> functions
      can be used for retrieving system load values, and
      the <c>util/0</c> and <c>util/1</c> functions can be used for
      retrieving CPU utilization values.</p>
    <p>When run on Linux, <c>cpu_sup</c> assumes that the <c>/proc</c>
      file system is present and accessible by <c>cpu_sup</c>. If it is
      not, <c>cpu_sup</c> will terminate.</p>
  </description>
  <funcs>
    <func>
      <name>nprocs() -> UnixProcesses | {error, Reason}</name>
      <fsummary>Get the number of UNIX processes running on this host</fsummary>
      <type>
        <v>UnixProcesses = int()</v>
        <v>Reason = term()</v>
      </type>
      <desc>
        <p>Returns the number of UNIX processes running on this machine.
          This is a crude way of measuring the system load, but it may
          be of interest in some cases.</p>
        <p>Returns 0 if <c>cpu_sup</c> is not available.</p>
      </desc>
    </func>
    <func>
      <name>avg1() -> SystemLoad | {error, Reason}</name>
      <fsummary>Get the system load average for the last minute</fsummary>
      <type>
        <v>SystemLoad = int()</v>
        <v>Reason = term()</v>
      </type>
      <desc>
        <p>Returns the average system load in the last minute, as
          described above. 0 represents no load, 256 represents the load
          reported as 1.00 by <c>rup</c>.</p>
        <p>Returns 0 if <c>cpu_sup</c> is not available.</p>
      </desc>
    </func>
    <func>
      <name>avg5() -> SystemLoad | {error, Reason}</name>
      <fsummary>Get the system load average for the last five minutes</fsummary>
      <type>
        <v>SystemLoad = int()</v>
        <v>Reason = term()</v>
      </type>
      <desc>
        <p>Returns the average system load in the last five minutes, as
          described above. 0 represents no load, 256 represents the load
          reported as 1.00 by <c>rup</c>.</p>
        <p>Returns 0 if <c>cpu_sup</c> is not available.</p>
      </desc>
    </func>
    <func>
      <name>avg15() -> SystemLoad | {error, Reason}</name>
      <fsummary>Get the system load average for the last fifteen minutes</fsummary>
      <type>
        <v>SystemLoad = int()</v>
        <v>Reason = term()</v>
      </type>
      <desc>
        <p>Returns the average system load in the last 15 minutes, as
          described above. 0 represents no load, 256 represents the load
          reported as 1.00 by <c>rup</c>.</p>
        <p>Returns 0 if <c>cpu_sup</c> is not available.</p>
      </desc>
    </func>
    <func>
      <name>util() -> CpuUtil | {error, Reason}</name>
      <fsummary>Get the CPU utilization</fsummary>
      <type>
        <v>CpuUtil = float()</v>
        <v>Reason = term()</v>
      </type>
      <desc>
        <p>Returns CPU utilization since the last call to
          <c>util/0</c> or <c>util/1</c> by the calling process.</p>
        <note>
          <p>The returned value of the first call to <c>util/0</c> or
            <c>util/1</c> by a process will on most systems be the CPU
            utilization since system boot, but this is not guaranteed
            and the value should therefore be regarded as garbage. This
            also applies to the first call after a restart of
            <c>cpu_sup</c>.</p>
        </note>
        <p>The CPU utilization is defined as the sum of the percentage
          shares of the CPU cycles spent in all busy processor states
          (see <c>util/1</c> below) in average on all CPUs.</p>
        <p>Returns 0 if <c>cpu_sup</c> is not available.</p>
      </desc>
    </func>
    <func>
      <name>util(Opts) -> UtilSpec | {error, Reason}</name>
      <fsummary>Get the CPU utilization</fsummary>
      <type>
        <v>Opts = [detailed | per_cpu]</v>
        <v>UtilSpec = UtilDesc | [UtilDesc]</v>
        <v>&nbsp;UtilDesc = {Cpus, Busy, NonBusy, Misc}</v>
        <v>&nbsp;&nbsp;Cpus = all | int() | [int()]()</v>
        <v>&nbsp;&nbsp;Busy = NonBusy = {State, Share} | Share</v>
        <v>&nbsp;&nbsp;&nbsp;State = user | nice_user | kernel</v>
        <v>&nbsp;&nbsp;&nbsp;&nbsp;| wait | idle | atom()</v>
        <v>&nbsp;&nbsp;&nbsp;Share = float()</v>
        <v>&nbsp;&nbsp;Misc = []</v>
        <v>Reason = term()</v>
      </type>
      <desc>
        <p>Returns CPU utilization since the last call to
          <c>util/0</c> or <c>util/1</c> by the calling process, in
          more detail than <c>util/0</c>.</p>
        <note>
          <p>The returned value of the first call to <c>util/0</c> or
            <c>util/1</c> by a process will on most systems be the CPU
            utilization since system boot, but this is not guaranteed
            and the value should therefore be regarded as garbage. This
            also applies to the first call after a restart of
            <c>cpu_sup</c>.</p>
        </note>
        <p>Currently recognized options:</p>
        <taglist>
          <tag><c>detailed</c></tag>
          <item>
            <p>The returned <c>UtilDesc</c>(s) will be even more
              detailed.</p>
          </item>
          <tag><c>per_cpu</c></tag>
          <item>
            <p>Each CPU will be specified separately (assuming this
              information can be retrieved from the operating system),
              that is, a list with one <c>UtilDesc</c> per CPU will be
              returned.</p>
          </item>
        </taglist>
        <p>Description of <c>UtilDesc = {Cpus, Busy, NonBusy, Misc}</c>:</p>
        <taglist>
          <tag><c>Cpus</c></tag>
          <item>
            <p>If the <c>detailed</c> and/or <c>per_cpu</c> option is
              given, this is the CPU number, or a list of the CPU
              numbers.</p>
            <p>If not, this is the atom <c>all</c> which implies that
              the <c>UtilDesc</c> contains information about all CPUs.</p>
          </item>
          <tag><c>Busy</c></tag>
          <item>
            <p>If the <c>detailed</c> option is given, this is a list
              of <c>{State, Share}</c> tuples, where each tuple
              contains information about a processor state that has
              been identified as a busy processor state (see below).
              The atom <c>State</c> is the name of the state, and
              the float <c>Share</c> represents the percentage share of
              the CPU cycles spent in this state since the last call to
              <c>util/0</c> or <c>util/1</c>.</p>
            <p>If not, this is the sum of the percentage shares of
              the CPU cycles spent in all states identified as busy.</p>
            <p>If the <c>per_cpu</c> is not given, the value(s)
              presented are the average of all CPUs.</p>
          </item>
          <tag><c>NonBusy</c></tag>
          <item>
            <p>Similar to <c>Busy</c>, but for processor states that
              have been identified as non-busy (see below).</p>
          </item>
          <tag><c>Misc</c></tag>
          <item>
            <p>Currently unused; reserved for future use.</p>
          </item>
        </taglist>
        <p>Currently these processor states are identified as busy:</p>
        <taglist>
          <tag><c>user</c></tag>
          <item>
            <p>Executing code in user mode.</p>
          </item>
          <tag><c>nice_user</c></tag>
          <item>
            <p>Executing code in low priority (nice) user mode.
              This state is currently only identified on Linux.</p>
          </item>
          <tag><c>kernel</c></tag>
          <item>
            <p>Executing code in kernel mode.</p>
          </item>
        </taglist>
        <p>Currently these processor states are identified as non-busy:</p>
        <taglist>
          <tag><c>wait</c></tag>
          <item>
            <p>Waiting. This state is currently only identified on
              Solaris.</p>
          </item>
          <tag><c>idle</c></tag>
          <item>
            <p>Idle.</p>
          </item>
        </taglist>
        <note>
          <p>Identified processor states may be different on different
            operating systems and may change between different versions
            of <c>cpu_sup</c> on the same operating system. The sum of
            the percentage shares of the CPU cycles spent in all busy
            and all non-busy processor states will always add up to
            100%, though.</p>
        </note>
        <p>Returns <c>{all,0,0,[]}</c> if <c>cpu_sup</c> is not
          available.</p>
      </desc>
    </func>
  </funcs>

  <section>
    <title>See Also</title>
    <p><seealso marker="os_mon_app">os_mon(3)</seealso></p>
  </section>
</erlref>