shorewall_code/docs/ScalabilityAndPerformance.xml

<?xml version="1.0" encoding="UTF-8"?>
<!DOCTYPE article PUBLIC "-//OASIS//DTD DocBook XML V4.2//EN"
"http://www.oasis-open.org/docbook/xml/4.2/docbookx.dtd">
<article>
  <!--$Id$-->

  <articleinfo>
    <title>Scalability and Performance</title>

    <authorgroup>
      <author>
        <firstname>Tom</firstname>

        <surname>Eastep</surname>
      </author>
    </authorgroup>

    <pubdate><?dbtimestamp format="Y/m/d"?></pubdate>

    <copyright>
      <year>2006</year>

      <year>2007</year>

      <holder>Thomas M. Eastep</holder>
    </copyright>

    <legalnotice>
      <para>Permission is granted to copy, distribute and/or modify this
      document under the terms of the GNU Free Documentation License, Version
      1.2 or any later version published by the Free Software Foundation; with
      no Invariant Sections, with no Front-Cover, and with no Back-Cover
      Texts. A copy of the license is included in the section entitled
      <quote><ulink url="GnuCopyright.htm">GNU Free Documentation
      License</ulink></quote>.</para>
    </legalnotice>
  </articleinfo>

  <section id="Intro">
    <title>Introduction</title>

    <para>The performance of the <emphasis role="bold">shorewall
    start</emphasis> and <emphasis role="bold">shorewall restart</emphasis>
    commands when using Shorewall-shell is a frequent topic of questions. This
    article attempts to explain the scalability issues involved and to offer
    some tips for reducing the time required to compile a Shorewall
    configuration and to execute the compiled script.</para>

    <para>Ultimately, the solution to these performance problems is to migrate
    to the use of Shorewall-perl if at all possible.</para>
  </section>

  <section id="Groups">
    <title>Host Groups</title>

    <para>In this article, we will use the term <firstterm>host
    group</firstterm> to refer to a set of IP addresses accessed through a
    particular interface. In a Shorewall configuration, there is one host
    group for:</para>

    <itemizedlist>
      <listitem>
        <para>Each entry in <filename>/etc/shorewall/interfaces</filename>
        that contains the name of a zone in the first column.</para>
      </listitem>

      <listitem>
        <para>Each entry in <filename>/etc/shorewall/hosts</filename>.</para>
      </listitem>
    </itemizedlist>

    <para>As you can see, each host group is associated with a single
    <firstterm>zone</firstterm>.</para>
  </section>

  <section id="GroupScale">
    <title>Scaling by Host Groups</title>

    <para>For each host group, it is possible to attempt connections to every
    other host group; and if the host group has the <emphasis
    role="bold">routeback</emphasis> option, then it is possible for
    connections to be attempted from the host group to itself. So if there are
    <emphasis role="bold">H</emphasis> host groups defined in a Shorewall
    configuration, then the number of unique pairs of (<emphasis>source host
    group</emphasis>, <emphasis>destination host group</emphasis>) is
    <emphasis role="bold">H</emphasis>*<emphasis role="bold">H</emphasis> or
    <emphasis role="bold">H</emphasis><superscript>2</superscript>. In other
    words, the number of combinations is the square of the number of host
    groups and increasing the number of groups from <emphasis
    role="bold">H</emphasis> to <emphasis role="bold">H</emphasis>+1 adds
    <emphasis role="bold">H</emphasis> + <emphasis role="bold">H</emphasis> +
    1 = 2<emphasis role="bold">H</emphasis> + 1 additional
    combinations.</para>
  </section>

  <section id="ZoneScale">
    <title>Scaling by Zones</title>

    <para>A similar scaling issue applies to Shorewall zones. If there are
    <emphasis role="bold">Z</emphasis> zones, then connections may be
    attempted from a given zone <emphasis
    role="bold">Z</emphasis><subscript>n</subscript> to all of the other zones
    (including to <emphasis role="bold">Z</emphasis><subscript>n</subscript>
    itself). Hence, the number of combinations is the square of the number of
    zones or <emphasis
    role="bold">Z</emphasis><superscript>2</superscript>.</para>
  </section>

  <section id="Shorewall">
    <title>Scaling within the Shorewall Code</title>

    <para>Shorewall is written entirely in Bourne Shell. While this allows
    Shorewall to run on a wide range of distributions (included embedded
    ones), the shell programming environment is not ideal for writing the
    compiler portion of Shorewall. As a consequence, the code must repeatedly
    perform sequential searches of lists. If a list has <emphasis
    role="bold">N</emphasis> elements and a sequential search is made for each
    of those elements, then the number of comparisons is 1 + 2 + 3 + .... +
    <emphasis role="bold">N</emphasis> = <emphasis role="bold">N</emphasis> *
    (<emphasis role="bold">N</emphasis> + 1 ) / 2. So again, we see order
    <emphasis role="bold">N</emphasis><superscript>2</superscript>
    scaling.</para>
  </section>

  <section id="Improving">
    <title>Improving Performance</title>

    <para>Achieving good performance boils down to three things:</para>

    <itemizedlist>
      <listitem>
        <para>Use a light-weight shell and fast hardware. Especially in the
        compiler, a light-weight shell such as <command>ash</command> or
        <command>dash</command> can provide considerable improvement over
        <command>bash</command>.</para>
      </listitem>

      <listitem>
        <para>With all of the order <emphasis
        role="bold">N</emphasis><superscript>2</superscript> scaling that is
        implicit in the problem being solved, it is vital to keep <emphasis
        role="bold">N</emphasis> small.</para>

        <itemizedlist>
          <listitem>
            <para>If you have a large number of interfaces, use wild-cards
            ("+") in <filename>/etc/shorewall/interfaces</filename> and
            <filename>/etc/shorewall/hosts</filename> to reduce the number of
            host groups.</para>
          </listitem>

          <listitem>
            <para>Combine host groups with similar firewall requirements into
            a single zone.</para>
          </listitem>
        </itemizedlist>
      </listitem>

      <listitem>
        <para>Use NONE policies whereever appropriate. This helps especially
        in the rules activation phase of both script compilation and
        execution.</para>
      </listitem>
    </itemizedlist>

    <para>So while it is tempting to create lots of zones through entries in
    <filename>/etc/shorewall/hosts</filename>, such configurations
    <emphasis>always</emphasis> perform badly. In these cases, it is much
    better to have more rules than more zones because the performance scales
    linearly with the number of rules whereas it scales geometrically with the
    number of zones.</para>

    <para>Another tip worth noting has to do with the use of shell
    variables.</para>

    <para>Suppose that the following appears in
    <filename>/etc/shorewall/params</filename>:</para>

    <programlisting>HOSTS=&lt;ip1&gt;,&lt;ip2&gt;,&lt;ip3&gt;,...&lt;ipN&gt;</programlisting>

    <para>and suppose that $HOSTS appears in the SOURCE column of <emphasis
    role="bold">M</emphasis> ACCEPT rules. That would generate a total of
    <emphasis role="bold">N</emphasis> * <emphasis role="bold">M</emphasis>
    iptables ACCEPT rules.</para>

    <para>The number of rules can be reduced significantly by using an <ulink
    url="Actions.html">action</ulink>. Consider the following:</para>

    <blockquote>
      <para><filename>/etc/shorewall/actions</filename>:</para>

      <programlisting>AcceptHosts</programlisting>

      <para><filename>/etc/shorewall/action.AcceptHosts</filename>:</para>

      <programlisting>#TARGET SOURCE          DEST            PROTO   DEST    SOURCE     ORIGINAL     RATE
#                                               PORT    PORT(S)    DEST         LIMIT
ACCEPT  $HOSTS </programlisting>
    </blockquote>

    <para>If the <emphasis role="bold">M</emphasis> ACCEPT rules are now
    replaced with <emphasis role="bold">M</emphasis> AcceptHosts rules, the
    total number of rules will be <emphasis role="bold">N</emphasis> +
    <emphasis role="bold">M</emphasis>.</para>

    <para>Example (Accept net-&gt;fw SSH from $HOSTS):</para>

    <programlisting>#ACTION          SOURCE          DEST            PROTO   DEST    SOURCE          ORIGINAL        RATE            USER/
#                                                        PORT    PORT(S)         DEST            LIMIT           GROUP
AcceptHosts      net             $FW             tcp     22             </programlisting>
  </section>
</article>