Dependency management in distributed settings

Applications in a computation and communication saturated pervasive-computing environment have to adapt their behaviors according the context of the user and the physical space. These "context-aware" applications typically obtain high-level situational understanding by aggregating one or more outputs of distributed sensors. To reduce the complexity of developing and deploying these adaptive applications, we designed and implemented a flexible software infrastructure, Solar, for large-scale context fusion (Chen et al., 2004). Solar consists of a set of functionally equivalent nodes, named Planets, which peer together to form a service overlay. A sensor connects to a Planet to register a name with Solar's naming service and publish a stream of events, while an application connects to a Planet to subscribe to one or more sensor outputs by querying Solar's name space. In addition, applications may inject data-fusion operators into the Solar to reduce the computation and communication costs on the mobile devices. In this paper we present our approach to address the dependency management issues in Solar.