ConfigurationFilesModuleParameters.xml

   1 <?xml version='1.0' encoding='UTF-8'?><chapter xmlns="http://docbook.org/ns/docbook" xmlns:xl="http://www.w3.org/1999/xlink" version="5.0" xml:lang="en-US" xml:id="configurationfilesmoduleparameters">
   2     <title xml:id="configurationfilesmoduleparameters.title">Configuration Files and Module Parameters</title>
   3   <para>This section describes configuration files and module parameters and includes the following sections:</para>
   4   <itemizedlist>
   5     <listitem>
   6       <para><xref linkend="dbdoclet.50438293_15350"/></para>
   7     </listitem>
   8     <listitem>
   9       <para><xref linkend="dbdoclet.50438293_78010"/></para>
  10     </listitem>
  11   </itemizedlist>
  12   <section xml:id="dbdoclet.50438293_15350">
  13       <title>
  14           <indexterm><primary>configuring</primary></indexterm>
  15           <indexterm><primary>LNet</primary><see>configuring</see></indexterm>
  16
  17
  18           Introduction</title>
  19     <para>LNet network hardware and routing are now configured via module parameters. Parameters should be specified in the <literal>/etc/modprobe.d/lustre.conf</literal>file, for example:</para>
  20     <screen>options lnet networks=tcp0(eth2)</screen>
  21     <para>The above option specifies that this node should use the TCP protocol on the eth2 network interface.</para>
  22     <para>Module parameters are read when the module is first loaded. Type-specific LND modules (for instance, <literal>ksocklnd</literal>) are loaded automatically by the LNet module when LNet starts (typically upon <literal>modprobe ptlrpc</literal>).</para>
  23     <para>LNet configuration parameters can be viewed under <literal>/sys/module/lnet/parameters/</literal>, and LND-specific parameters under the name of the corresponding LND, for example <literal>/sys/module/ksocklnd/parameters/</literal> for the socklnd (TCP) LND.</para>
  24     <para>For the following parameters, default option settings are shown in parenthesis. Changes to parameters marked with a W affect running systems. Unmarked parameters can only be set when LNet loads for the first time.  Changes to parameters marked with <literal>Wc</literal> only have effect when connections are established (existing connections are not affected by these changes.)</para>
  25   </section>
  26   <section xml:id="dbdoclet.50438293_78010">
  27       <title>
  28           <indexterm><primary>configuring</primary><secondary>module options</secondary></indexterm>
  29
  30           Module Options</title>
  31     <itemizedlist>
  32       <listitem>
  33         <para>With routed or other multi-network configurations, use <literal>ip2nets</literal> rather than networks, so all nodes can use the same configuration.</para>
  34       </listitem>
  35       <listitem>
  36         <para>For a routed network, use the same &apos;routes&apos; configuration everywhere. Nodes specified as routers automatically enable forwarding and any routes that are not relevant to a particular node are ignored. Keep a common configuration to guarantee that all nodes have consistent routing tables.</para>
  37       </listitem>
  38       <listitem>
  39         <para>A separate <literal>lustre.conf</literal> file makes distributing the configuration much easier.</para>
  40       </listitem>
  41       <listitem>
  42         <para>If you set <literal>config_on_load=1</literal>, LNet starts at
  43             <literal>modprobe</literal> time rather than waiting for the Lustre file system to
  44           start. This ensures routers start working at module load time.</para>
  45       </listitem>
  46     </itemizedlist>
  47     <screen># lctl
  48 # lctl&gt; net down</screen>
  49     <itemizedlist>
  50       <listitem>
  51         <para>Remember the <literal>lctl ping {nid}</literal> command - it is a handy way to check your LNet configuration.</para>
  52       </listitem>
  53     </itemizedlist>
  54     <section remap="h3">
  55       <title><indexterm><primary>configuring</primary><secondary>LNet options</secondary></indexterm>
  56 LNet Options</title>
  57       <para>This section describes LNet options.</para>
  58       <section remap="h4">
  59         <title><indexterm><primary>configuring</primary><secondary>network topology</secondary></indexterm>
  60 Network Topology</title>
  61         <para>Network topology module parameters determine which networks a node should join, whether it should route between these networks, and how it communicates with non-local networks.</para>
  62         <para>Here is a list of various networks and the supported software stacks:</para>
  63         <informaltable frame="all">
  64           <tgroup cols="2">
  65             <colspec colname="c1" colwidth="50*"/>
  66             <colspec colname="c2" colwidth="50*"/>
  67             <thead>
  68               <row>
  69                 <entry>
  70                   <para><emphasis role="bold">Network</emphasis></para>
  71                 </entry>
  72                 <entry>
  73                   <para><emphasis role="bold">Software Stack</emphasis></para>
  74                 </entry>
  75               </row>
  76             </thead>
  77             <tbody>
  78               <row>
  79                 <entry>
  80                   <para> o2ib</para>
  81                 </entry>
  82                 <entry>
  83                   <para> OFED Version 2</para>
  84                 </entry>
  85               </row>
  86             </tbody>
  87           </tgroup>
  88         </informaltable>
  89         <note>
  90           <para>The Lustre software ignores the loopback interface (<literal>lo0</literal>), but the
  91             Lustre file system uses any IP addresses aliased to the loopback (by default). When in
  92             doubt, explicitly specify networks.</para>
  93         </note>
  94         <para><literal>ip2nets</literal> (&quot;&quot;) is a string that lists globally-available networks, each with a set of IP address ranges. LNet determines the locally-available networks from this list by matching the IP address ranges with the local IPs of a node. The purpose of this option is to be able to use the same <literal>modules.conf</literal> file across a variety of nodes on different networks. The string has the following syntax.</para>
  95         <screen>&lt;ip2nets&gt; :== &lt;net-match&gt; [ &lt;comment&gt; ] { &lt;net-sep&gt; &lt;net-match&gt; }
  96 &lt;net-match&gt; :== [ &lt;w&gt; ] &lt;net-spec&gt; &lt;w&gt; &lt;ip-range&gt; { &lt;w&gt; &lt;ip-range&gt; }
  97 [ &lt;w&gt; ]
  98 &lt;net-spec&gt; :== &lt;network&gt; [ &quot;(&quot; &lt;interface-list&gt; &quot;)&quot; ]
  99 &lt;network&gt; :== &lt;nettype&gt; [ &lt;number&gt; ]
 100 &lt;nettype&gt; :== &quot;tcp&quot; | &quot;elan&quot; | &quot;o2ib&quot; | ...
 101 &lt;iface-list&gt; :== &lt;interface&gt; [ &quot;,&quot; &lt;iface-list&gt; ]
 102 &lt;ip-range&gt; :== &lt;r-expr&gt; &quot;.&quot; &lt;r-expr&gt; &quot;.&quot; &lt;r-expr&gt; &quot;.&quot; &lt;r-expr&gt;
 103 &lt;r-expr&gt; :== &lt;number&gt; | &quot;*&quot; | &quot;[&quot; &lt;r-list&gt; &quot;]&quot;
 104 &lt;r-list&gt; :== &lt;range&gt; [ &quot;,&quot; &lt;r-list&gt; ]
 105 &lt;range&gt; :== &lt;number&gt; [ &quot;-&quot; &lt;number&gt; [ &quot;/&quot; &lt;number&gt; ] ]
 106 &lt;comment :== &quot;#&quot; { &lt;non-net-sep-chars&gt; }
 107 &lt;net-sep&gt; :== &quot;;&quot; | &quot;\n&quot;
 108 &lt;w&gt; :== &lt;whitespace-chars&gt; { &lt;whitespace-chars&gt; }
 109 </screen>
 110         <para><literal>&lt;net-spec&gt;</literal> contains enough information to
 111         uniquely identify the network and load an appropriate LND. The LND
 112         determines the missing &quot;address-within-network&quot; part of the
 113         NID based on the interfaces it can use.</para>
 114         <para><literal>&lt;iface-list&gt;</literal> specifies which hardware
 115         interface the network can use. If omitted, all interfaces are used. LNDs
 116         that do not support the <literal>&lt;iface-list&gt;</literal> syntax
 117         cannot be configured to use particular interfaces and just use what is
 118         there. Only a single instance of these LNDs can exist on a node at any
 119         time, and <literal>&lt;iface-list&gt;</literal> must be omitted.</para>
 120         <para><literal>&lt;net-match&gt;</literal> entries are scanned in the
 121         order declared to see if one of the node&apos;s IP addresses matches one
 122         of the <literal>&lt;ip-range&gt;</literal> expressions. If there is a
 123         match, <literal>&lt;net-spec&gt;</literal> specifies the network to
 124         instantiate. Note that it is the first match for a particular network
 125         that counts. This can be used to simplify the match expression for the
 126         general case by placing it after the special cases. For example:</para>
 127         <screen>ip2nets=&quot;tcp(eth1,eth2) 134.32.1.[4-10/2]; tcp(eth1) *.*.*.*&quot;</screen>
 128         <para>4 nodes on the 134.32.1.* network have 2 interfaces
 129         (134.32.1.{4,6,8,10}) but all the rest have 1.</para>
 130         <screen>ip2nets=&quot;<emphasis role="bold">o2ib</emphasis> 192.168.0.*; tcp(eth2) 192.168.0.[1,7,4,12]&quot; </screen>
 131         <para>This describes an IB cluster on 192.168.0.*. Four of these nodes
 132         also have IP interfaces; these four could be used as routers.</para>
 133         <para>Note that match-all expressions (For instance,
 134         <literal>*.*.*.*</literal>) effectively mask all other</para>
 135         <para> <literal>&lt;net-match&gt;</literal> entries specified after
 136         them. They should be used with caution.</para>
 137         <para>Here is a more complicated situation, the route parameter is
 138         explained below. We have:</para>
 139         <itemizedlist>
 140           <listitem>
 141             <para>Two TCP subnets</para>
 142           </listitem>
 143           <listitem>
 144             <para>One Elan subnet</para>
 145           </listitem>
 146           <listitem>
 147             <para>One machine set up as a router, with both TCP and Elan
 148             interfaces</para>
 149           </listitem>
 150           <listitem>
 151             <para>IP over Elan configured, but only IP will be used to label the
 152             nodes.</para>
 153           </listitem>
 154         </itemizedlist>
 155         <screen>options lnet ip2nets=â€tcp 198.129.135.* 192.128.88.98; \
 156         elan 198.128.88.98 198.129.135.3; \
 157         routes=&apos;cp 1022@elan # Elan NID of router; \
 158         elan  198.128.88.98@tcp # TCP NID of router  &apos;</screen>
 159       </section>
 160       <section remap="h4">
 161           <title><indexterm><primary>configuring</primary>
 162           <secondary>network</secondary><tertiary>tcp</tertiary></indexterm>
 163           networks (&quot;tcp&quot;)</title>
 164         <para>This is an alternative to &quot;<literal>ip2nets</literal>&quot;
 165         which can be used to specify the networks to be instantiated explicitly.
 166         The syntax is a simple comma separated list of
 167         <literal>&lt;net-spec&gt;</literal>s (see above). The default is only
 168         used if neither &apos;ip2nets&apos; nor &apos;networks&apos; is
 169         specified.</para>
 170       </section>
 171       <section remap="h4">
 172           <title><indexterm><primary>configuring</primary><secondary>network</secondary><tertiary>routes</tertiary></indexterm>
 173 routes (&quot;&quot;)</title>
 174         <para>This is a string that lists networks and the NIDs of routers that forward to them.</para>
 175         <para>It has the following syntax (<literal>&lt;w&gt;</literal> is one or more whitespace characters):</para>
 176         <screen>&lt;routes&gt; :== &lt;route&gt;{ ; &lt;route&gt; }
 177 &lt;route&gt; :== [&lt;net&gt;[&lt;w&gt;&lt;hopcount&gt;]&lt;w&gt;&lt;nid&gt;[:&lt;priority&gt;]{&lt;w&gt;&lt;nid&gt;[:&lt;priority&gt;]}</screen>
 178         <para>Note: the priority parameter was added in release 2.5.</para>
 179         <para>So a node on the network <literal>tcp1</literal> that needs to go through a router to get to the Elan network:</para>
 180         <screen>options lnet networks=tcp1 routes=&quot;elan 1 192.168.2.2@tcpA&quot;</screen>
 181         <para>The hopcount and priority numbers are used to help choose the best path between multiply-routed configurations.</para>
 182         <para>A simple but powerful expansion syntax is provided, both for target networks and router NIDs as follows.</para>
 183         <screen>&lt;expansion&gt; :== &quot;[&quot; &lt;entry&gt; { &quot;,&quot; &lt;entry&gt; } &quot;]&quot;
 184 &lt;entry&gt; :== &lt;numeric range&gt; | &lt;non-numeric item&gt;
 185 &lt;numeric range&gt; :== &lt;number&gt; [ &quot;-&quot; &lt;number&gt; [ &quot;/&quot; &lt;number&gt; ] ]</screen>
 186         <para>The expansion is a list enclosed in square brackets. Numeric items in the list may be a single number, a contiguous range of numbers, or a strided range of numbers. For example, <literal>routes=&quot;elan 192.168.1.[22-24]@tcp&quot;</literal> says that network <literal>elan0</literal> is adjacent (hopcount defaults to 1); and is accessible via 3 routers on the <literal>tcp0</literal> network (<literal>192.168.1.22@tcp</literal>, <literal>192.168.1.23@tcp</literal> and <literal>192.168.1.24@tcp</literal>).</para>
 187         <para><literal>routes=&quot;[tcp,o2ib] 2 [8-14/2]@elan&quot;</literal>
 188 says that 2 networks (<literal>tcp0</literal> and <literal>o2ib0</literal>) are accessible through 4 routers (<literal>8@elan</literal>, <literal>10@elan</literal>, <literal>12@elan</literal> and <literal>14@elan</literal>). The hopcount of 2 means that traffic to both these networks will be traversed 2 routers - first one of the routers specified in this entry, then one more.</para>
 189         <para>Duplicate entries, entries that route to a local network, and entries that specify routers on a non-local network are ignored.</para>
 190         <para>Prior to release 2.5, a conflict between equivalent entries was resolved in favor of the route with the shorter hopcount. The hopcount, if omitted, defaults to 1 (the remote network is adjacent)..</para>
 191         <para condition='l25'>Since 2.5, equivalent entries are resolved in favor of the route with the lowest priority number or shorter hopcount if the priorities are equal. The priority, if omitted, defaults to 0.  The hopcount, if omitted, defaults to 1 (the remote network is adjacent).</para>
 192         <para>It is an error to specify routes to the same destination with routers on different local networks.</para>
 193         <para>If the target network string contains no expansions, then the hopcount defaults to 1 and may be omitted (that is, the remote network is adjacent). In practice, this is true for most multi-network configurations. It is an error to specify an inconsistent hop count for a given target network. This is why an explicit hopcount is required if the target network string specifies more than one network.</para>
 194       </section>
 195       <section remap="h4">
 196           <title><indexterm><primary>configuring</primary><secondary>network</secondary><tertiary>forwarding</tertiary></indexterm>
 197 forwarding (&quot;&quot;)</title>
 198         <para>This is a string that can be set either to &quot;<literal>enabled</literal>&quot; or &quot;<literal>disabled</literal>&quot; for explicit control of whether this node should act as a router, forwarding communications between all local networks.</para>
 199         <para>A standalone router can be started by simply starting LNet (&apos;<literal>modprobe ptlrpc</literal>&apos;) with appropriate network topology options.</para>
 200         <informaltable frame="all">
 201           <tgroup cols="2">
 202             <colspec colname="c1" colwidth="50*"/>
 203             <colspec colname="c2" colwidth="50*"/>
 204             <thead>
 205               <row>
 206                 <entry>
 207                   <para><emphasis role="bold">Variable</emphasis></para>
 208                 </entry>
 209                 <entry>
 210                   <para><emphasis role="bold">Description</emphasis></para>
 211                 </entry>
 212               </row>
 213             </thead>
 214             <tbody>
 215               <row>
 216                 <entry>
 217                   <para> <literal>acceptor</literal></para>
 218                 </entry>
 219                 <entry>
 220                   <para>The acceptor is a TCP/IP service that some LNDs use to establish communications. If a local network requires it and it has not been disabled, the acceptor listens on a single port for connection requests that it redirects to the appropriate local network. The acceptor is part of the LNet module and configured by the following options:</para>
 221                   <itemizedlist>
 222                     <listitem>
 223                       <para><literal>secure</literal>  - Accept connections only from reserved TCP ports (below 1023).</para>
 224                     </listitem>
 225                     <listitem>
 226                       <para><literal>all</literal>  - Accept connections from any TCP port. </para>
 227                       <note>
 228                         <para>This is required for liblustre clients to allow connections on non-privileged ports.</para>
 229                       </note>
 230                     </listitem>
 231                     <listitem>
 232                       <para><literal>none</literal>  - Do not run the acceptor.</para>
 233                     </listitem>
 234                   </itemizedlist>
 235                 </entry>
 236               </row>
 237               <row>
 238                 <entry>
 239                   <para> <literal>accept_port</literal></para>
 240                   <para> <literal>(988)</literal></para>
 241                 </entry>
 242                 <entry>
 243                   <para>  Port number on which the acceptor should listen for connection requests. All nodes in a site configuration that require an acceptor must use the same port.</para>
 244                 </entry>
 245               </row>
 246               <row>
 247                 <entry>
 248                   <para> <literal>accept_backlog</literal></para>
 249                   <para> <literal>(127)</literal></para>
 250                 </entry>
 251                 <entry>
 252                   <para>Maximum length that the queue of pending connections may grow to (see listen(2)).</para>
 253                 </entry>
 254               </row>
 255               <row>
 256                 <entry>
 257                   <para> <literal>accept_timeout</literal></para>
 258                   <para> <literal>(5, W)</literal></para>
 259                 </entry>
 260                 <entry>
 261                   <para>Maximum time in seconds the acceptor is allowed to block while communicating with a peer.</para>
 262                 </entry>
 263               </row>
 264               <row>
 265                 <entry>
 266                   <para> <literal>accept_proto_version</literal></para>
 267                 </entry>
 268                 <entry>
 269                   <para>Version of the acceptor protocol that should be used by outgoing connection requests. It defaults to the most recent acceptor protocol version, but it may be set to the previous version to allow the node to initiate connections with nodes that only understand that version of the acceptor protocol. The acceptor can, with some restrictions, handle either version (that is, it can accept connections from both &apos;old&apos; and &apos;new&apos; peers). For the current version of the acceptor protocol (version 1), the acceptor is compatible with old peers if it is only required by a single local network.</para>
 270                 </entry>
 271               </row>
 272             </tbody>
 273           </tgroup>
 274         </informaltable>
 275       </section>
 276       <section>
 277           <title><indexterm><primary>configuring</primary><secondary>network</secondary><tertiary>rnet_htable_size</tertiary></indexterm>
 278 <literal>rnet_htable_size</literal></title>
 279         <para><literal>rnet_htable_size</literal> is an integer that indicates how many remote networks the internal LNet hash table is configured to handle. <literal>rnet_htable_size</literal> is used for optimizing the hash table size and does not put a limit on how many remote networks you can have.  The default hash table size when this parameter is not specified is: 128.</para>
 280       </section>
 281     </section>
 282     <section remap="h3" xml:id="section_ngq_qhy_zl">
 283       <title><indexterm>
 284           <primary>configuring</primary>
 285           <secondary>network</secondary>
 286           <tertiary>SOCKLND</tertiary>
 287         </indexterm>
 288         <literal>SOCKLND</literal> Kernel TCP/IP LND</title>
 289       <para>The <literal>SOCKLND</literal> kernel TCP/IP LND (<literal>socklnd</literal>) is
 290         connection-based and uses the acceptor to establish communications via sockets with its
 291         peers.</para>
 292       <para>It supports multiple instances and load balances dynamically over multiple interfaces.
 293         If no interfaces are specified by the <literal>ip2nets</literal> or networks module
 294         parameter, all non-loopback IP interfaces are used. The address-within-network is determined
 295         by the address of the first IP interface an instance of the <literal>socklnd</literal>
 296         encounters.</para>
 297       <para>Consider a node on the &apos;edge&apos; of an InfiniBand network,
 298       with a low-bandwidth management Ethernet (<literal>eth0</literal>), IP
 299       over IB configured (<literal>ipoib0</literal>), and a pair of GigE NICs
 300       (<literal>eth1</literal>,<literal>eth2</literal>) providing off-cluster
 301       connectivity. This node should be configured with &apos;
 302       <literal>networks=o2ib,tcp(eth1,eth2)</literal>&apos; to ensure that the
 303       <literal>socklnd</literal> ignores the management Ethernet and IPoIB.
 304       </para>
 305       <informaltable frame="all">
 306         <tgroup cols="2">
 307           <colspec colname="c1" colwidth="50*"/>
 308           <colspec colname="c2" colwidth="50*"/>
 309           <thead>
 310             <row>
 311               <entry>
 312                 <para><emphasis role="bold">Variable</emphasis></para>
 313               </entry>
 314               <entry>
 315                 <para><emphasis role="bold">Description</emphasis></para>
 316               </entry>
 317             </row>
 318           </thead>
 319           <tbody>
 320             <row>
 321               <entry>
 322                 <para>
 323                   <literal>timeout</literal></para>
 324                 <para>
 325                   <literal>(50,W)</literal></para>
 326               </entry>
 327               <entry>
 328                 <para>Time (in seconds) that communications may be stalled before the LND completes
 329                   them with failure.</para>
 330               </entry>
 331             </row>
 332             <row>
 333               <entry>
 334                 <para>
 335                   <literal>nconnds</literal></para>
 336                 <para>
 337                   <literal>(4)</literal></para>
 338               </entry>
 339               <entry>
 340                 <para>Sets the number of connection daemons.</para>
 341               </entry>
 342             </row>
 343             <row>
 344               <entry>
 345                 <para>
 346                   <literal>min_reconnectms</literal></para>
 347                 <para>
 348                   <literal>(1000,W)</literal></para>
 349               </entry>
 350               <entry>
 351                 <para>Minimum connection retry interval (in milliseconds). After a failed connection
 352                   attempt, this is the time that must elapse before the first retry. As connections
 353                   attempts fail, this time is doubled on each successive retry up to a maximum of
 354                     &apos;<literal>max_reconnectms</literal>&apos;.</para>
 355               </entry>
 356             </row>
 357             <row>
 358               <entry>
 359                 <para>
 360                   <literal>max_reconnectms</literal></para>
 361                 <para>
 362                   <literal>(6000,W)</literal></para>
 363               </entry>
 364               <entry>
 365                 <para>Maximum connection retry interval (in milliseconds).</para>
 366               </entry>
 367             </row>
 368             <row>
 369               <entry>
 370                 <para>
 371                   <literal>eager_ack</literal></para>
 372                 <para>
 373                   <literal>(0 on linux,</literal></para>
 374                 <para>
 375                   <literal>1 on darwin,W)</literal></para>
 376               </entry>
 377               <entry>
 378                 <para>Boolean that determines whether the <literal>socklnd</literal> should attempt
 379                   to flush sends on message boundaries.</para>
 380               </entry>
 381             </row>
 382             <row>
 383               <entry>
 384                 <para>
 385                   <literal>typed_conns</literal></para>
 386                 <para>
 387                   <literal>(1,Wc)</literal></para>
 388               </entry>
 389               <entry>
 390                 <para>Boolean that determines whether the <literal>socklnd</literal> should use
 391                   different sockets for different types of messages. When clear, all communication
 392                   with a particular peer takes place on the same socket. Otherwise, separate sockets
 393                   are used for bulk sends, bulk receives and everything else.</para>
 394               </entry>
 395             </row>
 396             <row>
 397               <entry>
 398                 <para>
 399                   <literal>min_bulk</literal></para>
 400                 <para>
 401                   <literal>(1024,W)</literal></para>
 402               </entry>
 403               <entry>
 404                 <para>Determines when a message is considered &quot;bulk&quot;.</para>
 405               </entry>
 406             </row>
 407             <row>
 408               <entry>
 409                 <para>
 410                   <literal>tx_buffer_size, rx_buffer_size</literal></para>
 411                 <para>
 412                   <literal>(8388608,Wc)</literal></para>
 413               </entry>
 414               <entry>
 415                 <para>Socket buffer sizes. Setting this option to zero (0), allows the system to
 416                   auto-tune buffer sizes. </para>
 417                 <warning>
 418                   <para>Be very careful changing this value as improper sizing can harm
 419                     performance.</para>
 420                 </warning>
 421               </entry>
 422             </row>
 423             <row>
 424               <entry>
 425                 <para>
 426                   <literal>nagle</literal></para>
 427                 <para>
 428                   <literal>(0,Wc)</literal></para>
 429               </entry>
 430               <entry>
 431                 <para>Boolean that determines if <literal>nagle</literal> should be enabled. It
 432                   should never be set in production systems.</para>
 433               </entry>
 434             </row>
 435             <row>
 436               <entry>
 437                 <para>
 438                   <literal>keepalive_idle</literal></para>
 439                 <para>
 440                   <literal>(30,Wc)</literal></para>
 441               </entry>
 442               <entry>
 443                 <para>Time (in seconds) that a socket can remain idle before a keepalive probe is
 444                   sent. Setting this value to zero (0) disables keepalives.</para>
 445               </entry>
 446             </row>
 447             <row>
 448               <entry>
 449                 <para>
 450                   <literal>keepalive_intvl</literal></para>
 451                 <para>
 452                   <literal>(2,Wc)</literal></para>
 453               </entry>
 454               <entry>
 455                 <para>Time (in seconds) to repeat unanswered keepalive probes. Setting this value to
 456                   zero (0) disables keepalives.</para>
 457               </entry>
 458             </row>
 459             <row>
 460               <entry>
 461                 <para>
 462                   <literal>keepalive_count</literal></para>
 463                 <para>
 464                   <literal>(10,Wc)</literal></para>
 465               </entry>
 466               <entry>
 467                 <para>Number of unanswered keepalive probes before pronouncing socket (hence peer)
 468                   death.</para>
 469               </entry>
 470             </row>
 471             <row>
 472               <entry>
 473                 <para>
 474                   <literal>enable_irq_affinity</literal></para>
 475                 <para>
 476                   <literal>(0,Wc)</literal></para>
 477               </entry>
 478               <entry>
 479                 <para>Boolean that determines whether to enable IRQ affinity. The default is zero
 480                   (0).</para>
 481                 <para>When set, <literal>socklnd</literal> attempts to maximize performance by
 482                   handling device interrupts and data movement for particular (hardware) interfaces
 483                   on particular CPUs. This option is not available on all platforms. This option
 484                   requires an SMP system to exist and produces best performance with multiple NICs.
 485                   Systems with multiple CPUs and a single NIC may see increase in the performance
 486                   with this parameter disabled.</para>
 487               </entry>
 488             </row>
 489             <row>
 490               <entry>
 491                 <para>
 492                   <literal>zc_min_frag</literal></para>
 493                 <para>
 494                   <literal>(2048,W)</literal></para>
 495               </entry>
 496               <entry>
 497                 <para>Determines the minimum message fragment that should be considered for
 498                   zero-copy sends. Increasing it above the platform&apos;s <literal>PAGE_SIZE
 499                   </literal>disables all zero copy sends. This option is not available on all
 500                   platforms.</para>
 501               </entry>
 502             </row>
 503           </tbody>
 504         </tgroup>
 505       </informaltable>
 506     </section>
 507   </section>
 508 </chapter>