Exploding the price-per-port fallacy

Mention a new LAN switch to a colleague and, almost without fail, the response will be, "What is its price per port?" So ingrained is this approach to comparing LAN switches that network managers and vendors rarely, if ever, question its validity. They should.

The belief that simple per-port pricing provides an accurate and valid switch cost comparison is a fallacy. Implicit in any comparison based on price is the assumption that the units are comparable. It doesn't take much to realize that an item that sells for $1 per liter in Germany is not cheaper than the same item selling for $2 per gallon in the U.S. The units are well-defined, and an accurate comparison price can be derived.

Unfortunately, the same cannot be said of the elusive switch port. What do you think you are buying when you purchase a switch with 24 full-duplex Fast Ethernet ports? Given that each port can, technically, transport 100M bit/sec in each direction, the answer has to be 24x100M bit/sec, or 2.4G bit/sec of total system bandwidth. Thus, when we compare three different brands of 24-port systems, what we are really comparing is the cost for 2.4G bit/sec of system bandwidth. When the systems being compared are not delivering the same level of system bandwidth, price per port becomes worse than irrelevant. It becomes misleading.

A recent test we conducted illustrates this. In best-case tests, the aggregate system throughput levels of the three products tested were 0.4G bit/sec, 1.4G bit/sec and 2.4G bit/sec, respectively. The per-port price of the highest performing product was roughly 20% to 30% higher than that of the lowest, yet the peak throughput was six times as great. With such a wide disparity in the effective utilization of each port, price per port is meaningless. More meaningful would be an examination of price-per-gigabit throughput as a way of determining value. Taking the retail prices as tested and the best-case results yields a cost-per-gigabit throughput of approximately $6,000, $2,600 and $1,500, respectively, for these switches.

I believe the numbers tell you much more about what you get for your money .For their part, though, vendors are not necessarily trying to deceive the buying public. Many claim, for example, that their switches are in different classes and should not be compared with products from other classes. Every one of us has heard the terms desktop, workgroup and backbone switches tossed around constantly. What is missing is a definition of each. I propose that the industry embrace a straightforward definition. Only products that can forward an aggregate of 75% or more of a device's total port capacity - at all frame sizes - should be called backbone switches. Workgroup switches would be those that fall into the 25% to 74% range. Any device under 25% would be advertised as a desktop switch.

Ironically, acceptance of a common definition that would provide for an empirical classification of LAN switches would actually help to validate price per port as a cost yardstick. After all, switches so classified would only be compared against devices that have similar performance characteristics. Given the ongoing importance of these devices, finding a valid comparison method is a concern that every network manager should share.