2-1 Abstract – This invited paper considers the challenges and opportunities in the provision of active processing in broadband networks. Active networking aims to provide a systematic mechanism to use processing capabilities within network nodes (switches and routers) to allow the dynamic provisioning and composition of advanced services. Many of these services are traditionally offered at the application layer or require changes to the network layer standards. Active networking has the potential for improving performance by providing functionality at the right place and layer, without the need to go up to the application layer. The processing gains achieved by Moore’s Law are frequently touted as the motivation for active networking. It is the ratio of processing and memory to bandwidth, however, that governs how much active processing is achievable. Active node architectures should support active processing with a flexible mix of software and programmable hardware based on the granularity of the processing. Significant practical challenges remain before active networking technology will be deployed and adopted. In particular, active networking is a technology awaiting the killer application. Index Terms – active high -speed broadband optical networking I. INTRODUCTION TO ACTIVE NETWORKING Active networking is a technology that uses processing in network nodes (switches and routers) to allow the dynamic provisioning and composition of enhanced services [1,18]. Many of these services are traditionally provided at the application layer (such as content caching), or require changes to the network layer standards (such as for multicast). Active networking has recently been a topic of intense research. In its current form, most of this research is centered on IP-based networking, motivated by the premise that processing and memory have become cheap enough to allow significant processing by network nodes (IP routers and switches). There is no fundamental reason, however, why active networking cannot be based on other network layer technologies, such as ATM. The implications of resource cost will be discussed in more detail in Section IV. This work was sponsored in part by the Defense Advanced Research Projects Agency under contract F30602-99-C-0131 issued by the AFRL. James P.G. Sterbenz is with BBN Technologies in Cambridge MA, 02138- 1191, USA, jpgs@ieee.org . First, this paper begins with a short summary of active network architecture. Various factors that influence performance include how active processing is invoked, the granularity of processing, the location of active processing in the network, and the architecture of the active node. A. Active Network Flavors Active networking can be divided into two major types: strong and moderate 1 . Strong AN is the type originally proposed in the context of active networking [17], in which users inject capsules into the netw ork that contain code to be executed in active nodes. Strong AN raises a number of significant policy, security, and resource concerns, to prevent active applications from eavesdropping or interfering with one another, as well as the normal passive packet forwarding. These concerns must have solid solutions before strong AN is viable outside the research community. Moderate AN is the case in which a service provider provisions code into network nodes to dynamically deploy new protocols and services, without the traditional software upgrade process. Note, however, that one of the services that can be provisioned is an interpreter, providing the equivalent of strong AN functionality when desired. The main difference is that moderate AN is intended primarily as a way for network providers to monitor, control, manage, deploy, and evolve network protocols and services, whereas strong AN is a way for end users to deploy and execute active services. Moderate AN is somewhat less risky in terms of security and denial of service, and gives the network provider more control over what active services are deployed and how they are executed. B. Active Processing Granularity A significant determinant of the amount of processing required for active networking services is the granularity of processing. This affects how often active processing must occur, ranging from per bit, through per byte, word, cell, frame, burst of frames, to per flow or connection. Four ranges of granularity are important. 1 The term strong AN and moderate AN were coined by Bhattacharjee and Sterbenz, respectively in 1997. Intelligence in Future Broadband Networks: Challenges and Opportunities in High-Speed Active Networking James P.G. Sterbenz 0 -7803 -7257-3/02/$10.00 ©2002 IEEE.