Model-centric development of highly available software systems

In today's rapidly evolving marketplace, the ability to quickly build and deploy new systems is an increasingly critical factor in a company's success. For certain domains, such as telecommunications, it is taken for granted that systems will be highly available, with expectations of "5 9s" or even higher availability, translating to five minutes or less downtime per year. However, building highly available systems is generally very challenging, and becoming even more challenging as the systems increase in complexity. High availability (HA) middleware solutions partially address this challenge by providing common HA services that system developers can use. However, developers still need to spend significant effort integrating their systems with the HA middleware. In this paper, we present the Aurora Management Workbench (AMW) as a solution to the integration problem. AMW is an HA middleware and tools for building highly available distributed software systems. It is unique in its approach for developing highly available systems: developers focus only on describing key architectural abstractions of their system as well as system high availability needs in the form of a model. Tools then use the model to generate much of the code needed to integrate the system with the AMW HA middleware, which also uses the model to coordinate and control HA services at run-time. This paper describes our approach and our initial successes using it to develop commercial telecom systems.