Just What Could Possibly Go Wrong In B2B Integration?

One important trend in enterprise-scale IT has been the increasing use of business-to-business integration (B2Bi) technologies to automate business processes that cross organisational boundaries, such as the interactions between partner companies along the supply chain. It is relatively easy to describe a pattern of interaction, or choreography, in the case where everything proceeds smoothly. However, the abnormal cases, such as where a process fails or a message is lost, are much more complicated, and risk introducing data and process inconsistencies into computer-based systems. Current B2Bi technologies do not supply an infrastructure that can provide reliability without considerable sophistication from the architects and developers, since performance and organisational autonomy rule out using traditional ACID transactions which ensure consistent data in OLTP systems. As a first step towards guiding architects to the design of B2Bi systems that maintain consistency despite failures, this paper describes a variety of types of failure that can arise in practice, based on a realistic e-procurement scenario. We describe these failures in terms of the different types of state that naturally occur within the distributed system. Different types of failure can lead to different types of inconsistencies if they are not appropriately handled. Understanding the types of failure that need to be handled, or prevented, is essential to an architect or developer who must design and write handlers for all the exceptions that can occur in their workflows.