Distributed and Collaborative Design of Composite Service Deployments Using Graph Transformations

The design of composite service deployments (including software, hardware, network and storage), is complicated due to multiple cross cutting concerns, both functional and non-functional, and organizational division of expertise to multiple overlapping domains. In [1], we proposed an approach for deployment design that includes the usage of an abstract requirement graph, and its refinement into a concrete (physical) “desired state” deployment topology using a set of fine grained transformation rules expressing configuration knowledge, best practices and policy. The approach reduces the amortized complexity of the deployment process, and the associated risk, and is suitable for a distributed deployment design activity, where multiple domain experts collaborate, which is the reality in large enterprises. IBM Rational’s deployment modeling platform is based on this approach.

In this paper, we lay out the theoretical foundations of the graph transformation approach for distributed design of composite service deployments. We propose a formal model for deployment design based on the Double Push Out (DPO) graph transformation technique. Our formal model includes a configuration domain containing abstract and concrete graph classes, and a configuration framework associating a graph transformation set with a configuration domain. We formally define what it means for a configuration framework to satisfy correctness, completeness and convergence. We demonstrate the approach on an example configuration framework (in the area of communication networks). We prove that the example configuration framework satisfies correctness and convergence, weak completeness, but not completeness.