Video distribution over the Internet poses many challenges. Due to the best-effort nature of today's public data networks, end system applications cannot rely on either bandwidth or delay guarantees. We designed and implemented a prototype of a multicast video distribution architecture involving knowledgeable active routers, a scalable video codec based on the wavelet transformation, and a high-performance video scaling algorithm implemented as a router plug-in. The plug-in scales the video with an average overhead of only 22 /spl mu/s per video datagram and is installed on-the-fly on the routers after the sender starts transmitting video for the first time. Through experiments on our test network, we show that we can dramatically improve the video quality on the receivers (up to 15 dB PSNR) by scaling the video on the routers to almost any target bandwidth. The target bandwidth is evaluated by the router solely based on monitoring of the load situation of the router's downstream links and can be adjusted within 50 ms.

Present-day Internet protocol routers typically employ monolithic operating systems that are not easily upgradable and extensible. With the rapid rate of protocol development it is becoming increasingly important to dynamically upgrade router software in an incremental fashion. We have designed and implemented a high-performance, modular, extended services router software architecture in the Net BSD operating system kernel. This architecture allows code modules, called plugins, to be dynamically added and configured at run time. One of the novel features of our design is the ability to bind different plugins to individual flows; this allows for distinct plugin implementations to seamlessly coexist in the same runtime environment. We achieve high performance through a carefully designed modular architecture, an innovative packet classification algorithm that is highly efficient, and by caching that exploits the flow-like characteristics of Internet traffic. Compared to a monolithic best effort kernel, our implementation requires an average increase in packet processing overhead of only 8%, or 600 cycles per packet when running on an Intel Pentium Pro at 233 MHz. By shortcutting the forward loop based on the per-flow state we establish, we can forward packets up to three times faster than the best effort kernel.

Active networking allows the network infrastructure to be programmable. Previous research focused on two commonly separated approaches: "capsules" and "programmable switches". Capsules are typically small programs in packets which flow through the network and are executed in-band on nodes receiving them. Programmable switches are network devices which offer a back-door to inject code by a network administrator out-of-band in order to enhance the device's capabilities. By combining these two approaches, this paper proposes a novel system architecture which allows both application specific data processing in network nodes as well as rapid deployment of new network protocol implementations. Instead of carrying code, data packets carry pointers to digitally signed active modules initially loaded on-the-fly, inband from trusted code servers on the network. Packet processing runs at high speed, may access and modify the whole network subsystem and no potentially slow virtual machines are needed.

Present day routers typically employ monolithic operating systems which are not easily upgradable and extensible. With the rapid rate of protocol development it is becoming increasingly important to dynamically upgrade router software in an incremental fashion. We have designed and implemented a high performance, modular, extended integrated services router software architecture in the NetBSD operating system kernel. This architecture allows code modules, called plugins, to be dynamically added and configured at run time. One of the novel features of our design is the ability to bind different plugins to individual flows; this allows for distinct plugin implementations to seamlessly coexist in the same runtime environment. High performance is achieved through a carefully designed modular architecture; an innovative packet classification algorithm that is both powerful and highly efficient; and by caching that exploits the flow-like characteristics of Internet traffic. Compared to a monolithic best-effort kernel, our implementation requires an average increase in packet processing overhead of only 8%, or 500 cycles/2.1ms per packet when running on a P6/233.