Improving scalability of data-centric services using in-network traffic inspection

A growing number of network services are being constructed using the XML Web services architecture. By design, client interactions with such services are governed by open protocols such as UDDI, WSDL, and SOAP that permit programmatic discovery and functionality invocation. Additionally, for a large number of currently deployed services, it is possible to interpret service requests as structured accesses against a (physical or virtual) database. These two trends suggest the possibility of building network intermediaries that can inspect traffic flowing between clients and services, infer models for service access patterns, and potentially improve service scalability by taking actions such as replication, request redirection, or admission control. This paper reports on our experience designing and implementing such a network intermediary architecture. Experiments on an emulated WAN using synthetic workloads show that our approach achieves significant performance improvements in client-perceived response time.