<?xml version="1.0"?>
<!--
 Licensed to the Apache Software Foundation (ASF) under one
 or more contributor license agreements.  See the NOTICE file
 distributed with this work for additional information
 regarding copyright ownership.  The ASF licenses this file
 to you 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.
-->

<document>
  <properties>
    <title>Lateral TCP Auxiliary Cache</title>
    <author email="pete@kazmier.com">Pete Kazmier</author>
    <author email="ASmuts@therealm.com">Aaron Smuts</author>
  </properties>

  <body>
    <section name="Lateral TCP Auxiliary Cache">
      <p>
        The TCP Lateral Auxiliary Cache is an optional plug in for the
        JCS.  It is primarily intended to broadcast puts and removals to
        other local caches, though it can also get cached objects.  It
        functions by opening up a <code>SocketServer</code> that
        listens to a configurable port and by creating
        <code>Socket</code> connections with other local cache
        <code>SocketServers</code>.  It can be configured to connect to
        any number of servers.
      </p>
      <p>
        If there is an error connecting to another server or if an error
        occurs in transmission, it will move into a recovery mode.  In
        recovery mode the TCP Lateral Auxiliary Cache will continue to
        communicate with healthy servers while it tries to restore the
        connection with the server that is in error.
      </p>
      <p>
        The cache hub communicates with a facade that implements a
        zombie pattern (balking facade) to prevent blocking.  Puts and
        removals are queued and occur synchronously in the background.
        Get requests are synchronous and can potentially block for a
        configurable interval if there is a communication problem.
      </p>
      <subsection name="Non-UDP Discovery Configuration">
        <p>
          The configuration is fairly straightforward and is done in the
          auxiliary cache section of the <code>cache.ccf</code>
          configuration file.  In the example below, I created a TCP
          Lateral Auxiliary Cache referenced by <code>LTCP</code>.  It
          connects to two servers defined in a comma separated list in
          the <code>TcpServers</code> attribute.  It listens to port
          <code>1110</code> and does <code>AllowGet</code>.
			Setting <code>AllowGet</code>
          equal to <code>false</code> would cause the auxiliary cache to
          return <code>null</code> from any get request.  In most cases this
          attribute should be set to <code>false</code>, since if the
          lateral caches were properly configured, the elements in one
          would be present in all.
        </p>
        <source><![CDATA[
jcs.auxiliary.LTCP=org.apache.jcs.auxiliary.lateral.socket.tcp.LateralTCPCacheFactory
jcs.auxiliary.LTCP.attributes=org.apache.jcs.auxiliary.lateral.socket.tcp.TCPLateralCacheAttributes
jcs.auxiliary.LTCP.attributes.TcpServers=localhost:1111,localhost:1112
jcs.auxiliary.LTCP.attributes.TcpListenerPort=1110
jcs.auxiliary.LTCP.attributes.AllowGet=true
        ]]></source>
        <p>
		  A mostly configurationless mode is available for the TCP
		  lateral cache if you use the <a href="LateralUDPDiscovery.html">UDP Discovery</a>
		   mechanism.
        </p>
      </subsection>
      <subsection name="Send Only Configuration">
        <p>
          You can configure the TCP lateral cache to operate
          in send only mode by setting the <code>Receive</code> attribute
          to false.  By default the receive attribute is true.
          When it is set to false, the lateral cache will not
          establish a socket server.
        </p>
        <p>
           Setting receive to false allows you to broadcast puts
           and removes, but not receive any.  This is useful for
           nodes of an application that produce data, but are not
           involved in data retrieval.
        </p>
        <p>
          The configuration below is the same as above, except the
          <code>Receive</code> attribute is set to false.  It also uses UDP
          discovery to find the servers, rather than listing them in the
          servers attribute.
        </p>
        <source><![CDATA[
jcs.auxiliary.LTCP=org.apache.jcs.auxiliary.lateral.socket.tcp.LateralTCPCacheFactory
jcs.auxiliary.LTCP.attributes=org.apache.jcs.auxiliary.lateral.socket.tcp.TCPLateralCacheAttributes
#jcs.auxiliary.LTCP.attributes.TcpServers=
jcs.auxiliary.LTCP.attributes.TcpListenerPort=1118
jcs.auxiliary.LTCP.attributes.UdpDiscoveryAddr=228.5.6.8
jcs.auxiliary.LTCP.attributes.UdpDiscoveryPort=6780
jcs.auxiliary.LTCP.attributes.UdpDiscoveryEnabled=true
jcs.auxiliary.LTCP.attributes.Receive=true
jcs.auxiliary.LTCP.attributes.AllowGet=false
jcs.auxiliary.LTCP.attributes.IssueRemoveOnPut=false
jcs.auxiliary.LTCP.attributes.FilterRemoveByHashCode=false
        ]]></source>
      </subsection>

      <subsection name="Potential Issues">
        <p>
          The TCP Lateral Auxiliary Cache can provide a high level of
          consistency but it does not guarantee consistency between
          caches.  A put for the same object could be issued in two
          different local caches. Since the transmission is queued, a
          situation could occur where the item put last in one cache is
          overridden by a put request from another local cache.  The two
          local caches could potentially have different versions of the
          same item.  Like most caches, this is intended for high get
          and low put utilization, and this occurrence would hint at
          improper usage.  The RMI Remote cache makes this situation a
          bit less likely to occur, since the default behavior is to
          remove local copies on put operations.  If either local cache
          needed the item put in the above situation, it would have to
          go remote to retrieve it.  Both local copies would have been
          expired and would end up using the same version, though it is
          possible that the version stored remotely would not be the
          last version created.  The OCS4J tries to implement a locking
          system to prevent this from occurring, but the locking system
          itself could suffer from similar problems (when granting locks
          from two roughly simultaneous lock requests) and it would
          create a significant burden on all the caches involved.  Since
          this situation would be extremely rare and is nearly
          impossible to solve practically, for now JCS will not offer
          any type of locking.
        </p>
      </subsection>
      <subsection name="Recent">
        <p>
          I added a <code>IssueRemoveOnPut</code> attribute that
          causes the lateral cache to remove an element from the
          cache rather than inserting it when a put.  This allows the local caches to
          dictate their own memory usage pattern.
        </p>
      </subsection>
    </section>
  </body>
</document>