Towards Transparent Architectural Decisions for Software Deployment

Abstract

The operation of large scale information systems requires investment into hardware infrastructure and causes running cost for keeping it in a productive state. This especially applies in an enterprise environment where also expenses for software licenses costs or penalties for downtime occur. The deployment of software influences these costs both in their amount and their composition. In order to optimize them a transparent view on these costs and the deployment is mandatory. In this paper, we present a conceptual model of deployment. The model is populated by reverse engineering of deployment descriptors but as well uses runtime traces and usage profiles. We envision ­ having both made explicit on an architectural level ­ a comprehensive decision making and optimization of software deployment. deployment view of the systems architecture to other views, such as structural or behavioral ones. When optimizing existing systems, deployment decisions already made have to be handled and balanced with potential migration effort. Historic data regarding operation characteristics and usage profile are available. We envision that having an integrated view on the relevant aspects of deployment facilitates conscious architectural decision making. The planning of migration or change effort is supposed to be more reliable. Our experiences in industry show a limited awareness of the influence of deployment on operation cost. Nevertheless, nearly all industry partners are able to tell a story where such a decision caused high cost (e.g., replacing a third party component to reduce license cost due to usage on multiple servers or enhancing network connections to locations to handle the traffic a new feature introduced). Decisions are typically made implicitly or have been deliberated once (at the time of the first release) and remained unchallenged since then (although the system evolved).