The deployment architecture that is represented in Figure 2–2 uses several strategies to meet the availability requirements of the reference configuration. Availability requirements fall into the two categories that are discussed in the following sections:
Service availability means that a service is available, even when a service provider fails. Service availability is generally achieved using multiple identically configured service instances (redundancy). Redundancy eliminates single points of failure (assuming that simultaneous failure of all instances is extremely unlikely). If one instance providing a service fails, another instance is available to take over. This mechanism is known as service failover.
Service failover is supported in the reference configuration through two mechanisms:
Load balancing. Load balancing uses redundant hardware and software components to distribute requests for a service among multiple component instances that provide the service. This redundancy provides greater capacity than would be possible with a single instance. This redundancy also means that if any one instance of a components fails, other instances are available to assume a heavier load. Depending on the latent capacity that is built into the deployment, a failure might not result in significant degradation of performance. Load balancing is used in all of the service modules in the reference configuration.
Directory Server multimaster replication. The preferred solution for Directory Server, this mechanism provides data that is crucial to the operation of the entire deployment. Multimaster replication is specifically designed to synchronize data between the two (or more) Directory Server instances shown in the deployment architecture. Multimaster replication is the simplest directory service failover implementation and is suitable for all but the highest-end deployments that need to support millions of users.
Session state availability means that data associated with a user session is not lost during a service failover. When a service failover occurs, the session state data that is stored by the failed instance is made available to the failover instance. This mechanism is known as session failover. The result is that the service failover is transparent to the user: the user will not be required to log in again or to restart a business operation.
Session failover is supported in the reference configuration through two mechanisms:
Access Manager session failover. Access Manager session information is created when a user is authenticated and stored in a replicated database. This database is shared by Access Manager instances and accessed through Message Queue. If an Access Manager instance fails, the load balancer routes all user requests to a failover instance (service failover). The failover instance retrieves session information from the shared database and maintains the session.
Portlet session failover. The JSR 168 portlet specification requires portlets to map state information to an HTTP session. If a web container supports highly available HTTP sessions, and if a Portal Server instance fails, the HTTP session state can be recovered by the failover instance. In the reference configuration, Portal Server is deployed in an Application Server cluster, in which High Availability Session Store (HADB) is used to store and replicate portlet session state. The failover instance retrieves session information from HADB and maintains the session.
Portlet session failover requires availability of Access Manager session state. An Access Manager failure could therefore interfere with portlet session failover, unless Access Manager session failover is also implemented.
When a user is successfully authenticated with Access Manager, the browser is redirected to a Portal Server instance. A portal desktop session is created on this instance and is mapped to the user's Access Manager session. This portal desktop session is used to track Portal Server specific information such as the user's merged display profile and provider properties. If a Portal Server instance fails, the desktop session is automatically re-created by using the user's display profile and attributes that are stored in the Access Manager's user session. However, provider properties that are stored in local memory are lost.