Aggregation for implicit invocations

Abstract

Implicit invocations are a popular mechanism for exchanging information between software components without binding these strongly. This decoupling is particularly important in distributed systems when interacting components are not known until runtime. In most realistic distributed systems though, components require some information about each other, be it only about their presence or their number. Runtime systems for implicit invocations--so-called publish/subscribe systems--are thus often combined with other systems providing such information. Given the variety of requirements for information about interacting components across applications, this paper proposes a generic augmentation of implicit invocations: rather than extending a given publish/subscribe API and system in order to convey a particular type of information across interacting components, we describe domain-specific joinpoints that can be used to advise application-level invocation routers-so-called brokers--used by publish/subscribe systems. This enables aggregation of application-specific information to and from components in a scalable manner. After presenting our domain-specific joinpoint model, we describe its implementation inside the REDS publish/subscribe middleware. The empirical evaluation of our approach shows that: (a) it outperforms external aggregation systems, by collecting and distributing information with a limited overhead; (b) the deployment of new functionalities has virtually no overhead, even if it occurs while the publish/subscribe system is running.