WebSphere eXtreme Scale (formerly Data Grid), Version 6.1
             Operating Systems: AIX, HP-UX, Linux, Solaris, Windows, z/OS


HA managed policy applied to partitioning

Partitions use a One of N policy with quorum enabled by default, which means partitions can only be activated when the majority of the possible cluster members are online, or in a state of quorum. Each partitioned Java™ 2 Platform, Enterprise Edition (J2EE) application creating a set of partitions (more details in next section) creates an entire set of HA managed groups. Each partition is an official HA manager group and can be managed separately from other groups. Thus, the capability exists to mix and match partition policies differently for each application or individual partitions. Additionally, when the specifics are discussed, each partitioned J2EE application can actually subclassify portions of the overall partition set into subgroups, and manage them uniquely.

For example, assume a user is creating a stock trading application. They are looking to have one partitioned J2EE application that handles all stock types, but they wish to treat S&P 500 stocks differently due to trade volume (load) characteristics. When the application servers start, all partitions will be activated on the set of servers available at the time quorum is established. At this time, the administrator can set a new policy and balance the load more effectively using the HA manager's policy infrastructure.

In this example, the administrator wants to fairly balance the partitions across the entire cluster based on expected transaction volume. A policy can be created that will balance the S&P 500 partitions uniformly over all the existing cluster members and then balance all the other stocks across the same set of cluster members similarly. This approach guarantees that the stocks transactions against the S&P companies will be balanced across the entire cluster, versus randomly balancing all stocks over the cluster members. If the partitions were managed as one grouping, as in this case, the result could be that some cluster members may have an inordinate number of high volume S&P 500 related partitions and, thus, many more transactions than other servers. In addition, some servers may have a large number of stocks that receive little to no transaction volume in a day and are under utilized.

Other examples that utilize the HA manager policy support are to set preferred servers for specific partitions, predefine servers to be used for failover scenarios, or define whether a partitions should be sent back to the original server once the server is back online. Many other options are available and described in subsequent sections.




Related concepts
HA manager and partitioning facility
Concept topic    

Terms of Use | Feedback

Last updated: Oct 30, 2009 6:09:57 PM EDT
http://publib.boulder.ibm.com/infocenter/wxdinfo/v6r1/index.jsp?topic=/com.ibm.websphere.dataint.doc/info/WPF51/cwpfhapartition_pdf.html