Large software systems require decompositionalmechanisms in order tomake them tractable. Traditionally, MILs and IDLs have played this role by providing notations based on definition/ use bindings. In this paper we argue that current MIL/IDLs based on definition/use have some serious drawbacks. A significant problem is that they fail to distinguish between "implementation" and "interaction" relationships between modules. We propose an alternative model in which components interact along

more »

... ined lines of communication – or connectors. Connectors are defined as protocols that capture the expected patterns of communication between modules. We show how this leads to a scheme that is much more expressive for architectural relationships, that allows the formal definition of module interaction, and that supports its own form of automated checks and formal reasoning.