Frederic Herrmann

On the use of the SA forum checkpoint and AMF services

In an environment where hardware and software errors may take an application down, high levels of service availability can be obtained by combining redundant hardware infrastructure with powerful application management frameworks which are able to detect these errors and recover from them. The Service Availability Forum (SAF) Application Interface Specification (AIS) defines interfaces used by the application management framework (AMF) to control applications placed under their supervision. The main goal of the AMF is to guarantee that for a given service there is always at least one instance of an application providing the service. In most cases the application implementing a service also encapsulates some data which is critical for the availability of the service. This data (called the application state) has to be preserved in a way that if the application providing the service fails, the replacing application can restore the application state and continue providing the service with minimal service disruption. The SAF AIS defines interfaces (named checkpoint interfaces) by which the application can save and restore its state. This paper describes the use of SAF AMF and checkpoint services in implementing applications providing highly-available services. The use of AMF and checkpoint services for implementing applications with the 2N or N-way active redundancy models is presented in the paper. It briefly discusses the interactions between the application, and AMF/checkpoint services during both (i) fault-free operations, as well as (ii) the error recovery procedures. The paper also suggests the most suitable checkpoint options for both of these redundancy models, depending on the tradeoff between protecting the integrity of the application state and the overhead of state saving/restoration by the application.

Platform management with SA forum and its role to achieve high availability

With its next release, SA Forum will provide a new platform management service and so will close the gap between HPI and AIS services. This new service will match the configured system to the actually present hardware and will make hardware states easily accessible for software via the information model. At the same time it extends the information model with objects that represent operating systems and virtualization layers. The new service provides a state model for all these objects and APIs to inform its users about changes, such as operator interactions with the equipment. Administrative commands can be used for platform maintenance. The complete information model will allow fault correlation over all platform layers up to the application. This is necessary for proper fault analysis and to really achieve high availability.