Providing Open Architecture High Availability Solutions

4.5 HA Configuration and Cluster Management

Configuration and cluster management middleware is the key controlling entity in an HA

configuration and is implemented by HA middleware distributed amongst the HA cluster systems.

It maintains a system model of the components that comprise the cluster, defines how faults are

detected in the cluster and what action should be taken as a result. When failures occur it takes the

appropriate actions to reconfigure the cluster and notify and/or restart affected parts of the

application.

The HA cluster configuration and management must be able to operate in a heterogeneous cluster.

The member systems in the cluster may be of different types with different operating

environments, and may have HA management middleware from different suppliers. The system

model, the message protocols between cluster members, and application APIs must allow proper

heterogeneous interoperability.

4.6 Data Replication and Data Integrity

One of the most difficult aspects of HA application design can be the preservation of the dynamic

data and context despite component failures. Not only does it affect the design and testing

complexity, but it may also have significant impacts on the system’s performance, dependability

and accuracy.

An application usually stores long term data and context in a file system or database, but often has

other more transient data that represents the instantaneous context of dialogues and/or transactions.

The ability to efficiently replicate this kind of data can dramatically improve the application

recovery time.

The open HA framework needs to provide for at least some of the following:

• An HA file system native to the operating system environment

• An HA database system

• An HA shared memory subsystem

• An application context replication and recovery subsystem

The HA file system, database and shared memory need to be integrated with the overall HA

configuration and management subsystem so that their operation can be coordinated with the HA

reconfiguration activities (e.g., failure, repair and recovery handling).

Additionally the API to a file system or database may have additional or modified calls and returns

in an HA version when compared to a non-HA version.