Alien: a generalized computing model of active networks
David S. Alexander, Jonathan M. Smith · Scholarly Commons (University of Pennsylvania) · 1998
Active Networks are networks whose actions may be changed during operation either by introducing new code into the routers (active extensions) or by executing code contained in a packet passing through the switch (active packets). This ability to make changes is expected to ease deployment of new or modified protocols, allow support for uncommon protocols, ease network experimentation, and improve performance in those situations in which quick response to local information is crucial. To maintain security of the system, there must be some restrictions on what code can be executed by any given user. In ALIEN, we have taken the Caml language and restricted it. On this base, we have architected and built a system capable of loading active extensions and executing active packets. We have used this system to measure performance characteristics of each approach. We built two primary experiments: the Active Bridge and the Secure Active Network Environment (SANE). The Active Bridge implements a (mostly) 802.1D-compliant network bridge which is built up in layers through the use of active extensions. It can achieve a throughput of approximately 60Mbps on a 100Mbps Ethernet. SANE studies a number of questions including the performance and bottlenecks associated with the use of capsules. With this experiment, we achieve a throughput of 13Mbps. These experiments help to demonstrate that we must now add language costs to the traditional network per-packet and per-byte costs when modelling performance. We demonstrate that the utility of Active Networks is driven primarily by the speed of the network. At low bandwidths (e.g., ISDN and cable modems), processing all the data in a packet is quite plausible. As the bandwidth increases, the available computation per byte decreases to the point that only per-packet processing such as routing can be performed. Beyond that, a point arises at which only a small fraction of packets can be processed, indicating the need for separate fast and slow paths through the active element with a need to ensure that most packets travel through the fast path.