Service-oriented architecture (SOA) promises an elegant model that can easily handle dynamic and heterogeneous environments such as pervasive computing systems. However, in reality, frequent failures of resource-poor and low-cost sensors greatly diminish the guarantees on reliability and availability offered and expected by SOA. To provide a framework for building fault-resilient service-oriented pervasive computing system, we present a solution that includes a virtual sensor framework and a high performance service composition solution based on: (1) WS-Pro, a probe-based performance-driven service composition architecture, and (2) an abstract service composition template (ASCT) approach. The concept of virtual sensor enhances the availability of services, while the service composition solution ensures the system can efficiently adapt to changes and failures in the environment. This approach also includes a novel probebased monitoring technique to actively collect performance data, and the use of an extended Petri net model, FPQSPN, to model performance of service composition.