Cluster-based network servers are a cost-effective platform for constructing  large-scale network services. However, service clustering presents significant  challenges to system designers. Many components of the clustering support are  often implemented using centralized or hierarchical architectures, in order to  meet desired performance, consistency, and policy requirements. The effectiveness of these approaches has been demonstrated under well-controlled environments, but customized enhancements are often needed to deal with transient component  failures and fluctuating service loads.  Peer-to-peer systems are distributed systems without any centralized control or hierarchical organization, and with functionally equivalent software running at each node. The primary goal of this project is to improve the robustness of cluster-based network services using peer-to-peer system design principles. Specifically, this effort exploits three main principles of peer-to-peer system design: functionally symmetric architecture, slow-scaling overhead, and non-determinism. This project covers various components in service clustering support, including service location lookup, membership management, cluster-wide resource management, and replication consistency support.  Techniques investigated in this project can greatly benefit network service designers, for whom it can offer significant improvements in system robustness. In a broader context, this research complements the ongoing efforts of building global-scale peer-to-peer systems and services. In addition, this project directly benefits systems-area instruction improvement at the PI's institution.