Of the different performance metrics to be considered, this section focuses on the packet-per-second switching capacity of the equiments. Packets per second is a fundamental performance metric for packet-based switching systems.

Unless an explicit packet-per-second performance value is given, one will have to be derived by taking average bandwidth numbers and assuming an average certain packet size. Packet sizes tend to follow the Ethernet sizes, which are 64 to 1522 bytes (this value includes the Ethernet encapsulation).

For data applications, the dominant protocol is TCP/IP. TCP generates an ACK for every packet received. Most of today’s implementations combine two ACKs together into one packet. ACK packets are 64 bytes in size and travel in the opposite direction of the original packet. The typical data traffic today in the downstream will be Web traffic, file transfer, or streaming media, meaning big packets in the downstream and small packets in the upstream.

The average downstream packet rate is determined by:

Result: PPS Data per Downstream Average = 80,000 / 8 / 500 = 20 pps

Result: PPS Data per Upstream Average = 32,000 / 8 / 64 = 63 pps

In formula 67, there are several times more packets in the upstream than in the downstream. Yet, the traffic model is assuming that large packets dominate the downstream and the upstream is dominated by ACKs. With two ACKs per packet, a more realistic model would be to set the packet-per-second rate in the upstream to be half of that in the downstream.

Result: PPS Data per CM Average = 1.5 * 80,000 / 8 / 500 = 30 pps

Result: PPS Data per CM Peak = 1.5 * 400,000 / 8 / 500 = 150 pps

Formula 68 has some practical advantages. It recognizes that when one direction of traffic is saturated, the other direction will not be. This is because the asymmetry of the data transfer is different from the asymmetry of the media.