Page History

etc...).

 Scalability in tiered applications means, e.g.

* a single web server directing traffic to multiple application servers

* many servers talking to a single database

* servers talking to many databases

* etc.

 
The need for scalability is not only driven by sheer capacity demand (mostly going along with requests to use low cost hardware to run services). It is very much linked to other non-functional requirements such as (high-) availability and maintainability.

This is the reason why understanding the various architecture patterns which are supported by a given application stack is crucial to deliver the right solution for a given business demand aka. requirement. This HOWTO guide talks about various options to "lay out" the physical server landscape in order to achieve certain non-functional requirements.

Most of the examples are taken from various threads on the community mailing list and replicated here in an abstract way.

h2.


h2. Single instance of OFBiz running with public IP

\[standard case, describe pros cons here\]

h2. Apache "in front of" OFBiz single instance (no load balancing)

[Setup|FAQ - Tips - Tricks - Cookbook - HowTo#Howtousemodproxyajp]

\[describe pros and cons\]

h2. Using Apache to load balance across multiple instances of OFBiz

The below is taken from mails by Brett Palmer on user@ofbiz.apache.org

 === mail 1 ===

Here is a configuration we use to do load balancing in ofbiz.

  Others may do
something else.

1. One machine is running an Apache Web server with mod_jk setup to
communicate with different instances of ofbiz.  The mod_jk is setup to load
balance between the different ofbiz instances.  Here is a document that
helps you get setup:

[http://tomcat.apache.org/connectors-doc/generic_howto/loadbalancers.html

]

2. Then we start multiple boxes running ofbiz server - all pointing to the

same database.

  The servers are setup to communicate using mod_jk's AJP

protocol.

   You need to set the "jvm-route" in the

framework/base/config/ofbiz-containers.xml file to insure the load balancer

maintains sessions between requests.

  Here is a snippet of that
configuration file.

<container name="catalina-container"

class="org.ofbiz.catalina.container.CatalinaContainer">

<\!-\- static configuration for tomcat \-->

<property name="delegator-name" value="default"/>

<property name="use-naming" value="false"/>

<property name="debug" value="0"/>

<property name="catalina-runtime-home" value="runtime/catalina"/>

<property name="apps-context-reloadable" value="false"/>

<property name="apps-cross-context" value="false"/>

<property name="apps-distributable" value="false"/>

<\!-\- one or more tomcat engines (servers); map to this + host \-->

<property name="default-server" value="engine">

<property name="default-host" value="0.0.0.0"/>

\*<property name="jvm-route" value="jvm1"/>\*

<property name="access-log-pattern">

.....

=== /mail 1 ===

 === mail 2 ===

1.

  We run one instance of ofbiz on a separate server but this is because we
are running a single application and want to scale by adding more machines.

2.  It is easier to use the embedded ofbiz version rather than a standalone
tomcat with ofbiz placed as a standard WAR file.  From the Apache Web server
perspective there isn't any difference.  It is talking AJP protocol and
can't tell if it is an embedded version of tomcat or standalone.

3.  I'm not sure the reason some are running several version of ofbiz on the
same machine.  My guess is they have different customers or application and
don't need the extra hardware to service these applications.  We multiple
instances of ofbiz on the same machine but this is so we can have a dev,
testing, and production version running on the same machine.

For maximum scalability you would put 1 instance of ofbiz on a single
machine and load balance between these machines.  We are using Amazon's EC2
servers so it is easy for us to shutdown to a single server at night and
then start up several server during the days when the application is busy.

=== /mail2 ===

h2.  Multiple OFBiz instances talking to the same database
\\ \\

h2. Clustering several OFBiz  instances

Clustering is a feature since Tomcat 5.5 and since OFBiz uses even Tomcat 6.0, the on board clustering with catalina has improved even more.

 The configuration for clustering is comprised (and by default commented out) with


{OFBIZ_HOME}/framework/base/config/ofbiz-containers.xml


<\!-\- &nbsp; uncomment for cluster support \-->
<property name="default-server-cluster" value="cluster">
&nbsp; <property name="rep-valve-filter">
&nbsp;&nbsp;&nbsp; <property-value>.*\.gif;.*\.js;.*\.jpg;.*\.htm;.*\.html;.*\.txt;</property-value>

&nbsp; </property>

&nbsp; <property name="manager-class" value="org.apache.catalina.ha.session.DeltaManager"/>

&nbsp; <property name="debug" value="5"/>

&nbsp; <property name="replication-mode" value="org.apache.catalina.tribes.transport.bio.PooledMultiSender"/>
&nbsp; <property name="tcp-listen-host" value="auto"/>
&nbsp; <property name="tcp-listen-port" value="4001"/>
&nbsp; <property name="tcp-sector-timeout" value="100"/>
&nbsp; <property name="tcp-thread-count" value="6"/>
&nbsp; <property name="mcast-bind-addr" value="172.17.88.52"/>
&nbsp; <property name="mcast-addr" value="228.0.0.4"/>
&nbsp; <property name="mcast-port" value="45564"/>
&nbsp; <property name="mcast-freq" value="500"/>
&nbsp; <property name="mcast-drop-time" value="3000"/>
</property>


h2. &nbsp;Multiple OFBiz instances on the same physical machine

The port configuration of each OFBiz instance is controlled in

{OFBIZ_HOME}/framework/base/config/ofbiz-containers.xml\\