Determine the following requirements for the application to be deployed.

• Required/acceptable response time.

• Peak load characteristics.

• Necessary persistence scope and frequency.

• Session timeout in web.xml.

• Failover and availability requirements.

  For more information see Oracle GlassFish Server 3.1 Performance Tuning Guide.

• Have necessary amounts of hard disk space and memory installed.

• Use the sizing exercise to identify the requirements for deployment.

  For more information see Oracle GlassFish Server 3.1-3.1.1 Release Notes

• Ensure that the product is installed on a supported platform.

• Ensure that the patch levels are up-to-date and accurate.

  For more information see Oracle GlassFish Server 3.1-3.1.1 Release Notes

• Identify single points of failures and address them.

• Make sure that the NICs and other network components are correctly configured.

• Run ttcp benchmark test to determine if the throughput meets the requirements/expected result.

• Setup ssh based your preference.

  For more information see Oracle GlassFish Server 3.1 Installation Guide.

Check with the domain expert or vendor to ensure that these data sources are configured appropriately.

• Make sure that changes to /etc/system and its equivalent on Linux are completed before running any performance/stress tests.

• Make sure the changes to the TCP/IP settings are complete.

• By default, the system comes with lots of services pre-configured. Not all of them are required to be running. Turn off services that are not needed to conserve system resources.

• On Solaris, use Setoolkit to determine the behavior of the system. Resolve any flags that show up.

  For more information see Oracle GlassFish Server 3.1 Performance Tuning Guide.

• Ensure that these servers are not installed on NFS mounted volumes.

• Check for enough disk space and RAM when installing GlassFish Server.

• Logging: Enable access log rotation.

• Choose the right logging level. WARNING is usually appropriate.

• Configure Java EE containers using the Administration Console.

• Configure HTTP listeners using the Administration Console.

• Configure ORB threadpool using the Administration Console.

• If using Type2 drivers or calls involving native code, ensure that mtmalloc.so is specified in the LD_LIBRARY_PATH.

• Ensure that the appropriate persistence scope and frequency are used and they are not overridden underneath in the individual Web/EJB modules.

• Ensure that only critical methods in the SFSB are checkpointed.

  For more information on tuning, see Oracle GlassFish Server 3.1 Performance Tuning Guide.

  For more information on configuration, see Oracle GlassFish Server 3.1 Administration Guide.

• Make sure the Web Server is installed.

• Make sure the load balancer plug-in into the Web Server is installed.

• Make sure patch checks is disabled.

• Lower the value of the KeepAliveQuery parameter. The lower the value, the lower the latency is on lightly loaded systems. The higher the value, the higher the throughput is on highly loaded systems.

• Initially set the minimum and maximum heap sizes to be the same, and at least one GB for each instance.

• See Java Hotspot VM Options for more information.

• When running multiple instances of GlassFish Server, consider creating a processor set and bind GlassFish Server to it. This helps in cases where the CMS collector is used to sweep the old generation.

• Response-time-out-in-seconds - How long the load balancer waits before declaring a GlassFish Server instance unhealthy. Set this value based on the response time of the application. If set too high, the Web Server and load balancer plug-in wait a long time before marking a GlassFish Server instance as unhealthy. If set too low and GlassFish Server’s response time crosses this threshold, the instance will be incorrectly marked as unhealthy.

• Interval-in-seconds - Time in seconds after which unhealthy instances are checked to find out if they have returned to a healthy state. Too low a value generates extra traffic from the load balancer plug-in to GlassFish Server instances and too high a value delays the routing of requests to the instance that has turned healthy.

• Timeout-in-seconds - Duration for a response to be obtained for a health check request. Adjust this value based on the traffic among the systems in the cluster to ensure that the health check succeeds.

• Max-wait-time-millis - Wait time to get a connection from the pool before throwing an exception. Default is 6 s. Consider changing this value for highly loaded systems where the size of the data being persisted is greater than 50 KB.

• Cache-idle-timeout-in-seconds - Time an EJB is allowed to be idle in the cache before it gets passivated. Applies only to entity beans and stateful session beans.

• Removal-timeout-in-seconds - Time that an EJB remains passivated (idle in the backup store). Default value is 60 minutes. Adjust this value based on the need for SFSB failover.

Garbage collection pauses of four seconds or more can cause intermittent problems in persisting session state. To avoid this problem, tune the VM heap. In cases where even a single failure to persist data is unacceptable or when the system is not fully loaded, use the CMS collector or the throughput collector.

These can be enabled by adding:

<jvm-options>-XX:+UseConcMarkSweepGC</jvm-options>

This option may decrease throughput.