Distributing Internet services to the network's edge

In the context of industrial information technology, the Internet and World Wide Web increasingly are seen as a solution to the problem of providing "anywhere, anytime" services. In the classical view of an Internet-enabled IT infrastructure, services are requested and consumed by a user (e.g., a human requesting plant production data from his or her desktop) and data are provided by an origin server (e.g., a Web server located in a plant that can authenticate users, implement encryption, serve data, and source multimedia streams). This rather simplistic view works well if the number of users is small, the complexity of services required is modest, and the real-time response requirements are lax. However, it fails to scale when one accounts for the complexities of modern networking: many simultaneous users, potentially operating in multiple languages; many complex data types, including incompatible display formats; many differing schemes for implementing privacy and security through many combinations of authentication and encryption. In this paper we propose an alternative-a client/edge server/origin server architecture that can distribute some complex data processing and device interface tasks to a network edge device, the NetEdge. We show how this device can support services thought to be useful to the industrial environment, such as language translation, image transcoding, access device adaptation, virus scanning, content assembly, local content insertion, and caching. The proposal is a win-win situation for all participants: industrial content providers need maintain only one copy of their content, yet consumers are provided with richer services and device-independent interfaces. Although the services provided define the utility of the product, the heart of the NetEdge is its rule engine. Rules specify which service requests, crossing specified processing points, invoke which service callouts. We explore how a proxylet interface connects the rule engine, through Java and C APIs, to the callout engine. We close with performance measurements of the NetEdge throughput and latency characteristics.