How is router 'capacity' conventionally defined?
How is router 'capacity' conventionally defined? I.E. is capacity equal to the N inputs and N outputs times the channel bit rate, or packets out of the router per second divided by the packets into the router per second?
Capacity is usually quoted as packets per second. The assumption is that routing processing is the bottleneck with software routers and with switches most of the processing is handled by hardware, which creates less of a bottleneck. This is usually for small IP packets, routed where feasible. Different routers have different performance curves for packet size, processing complexity. Cisco routers usually have several different processing paths, and some are more expensive per packet than others. Software version affects the performance. Other bottlenecks get in the way, such as memory, buses. Router architecture may alter performance (i.e. using VIPs in a 75xx.)
Cisco tends to give out headline numbers for box performance, but doesn't give out enough background information about test conditions. Just to complicate matters the figures are spread across lot of different documentation. What they are good at is relative performance numbers; for example the new 2620XM model is 30% "faster" than a 2620.
This was first published in August 2002