Put simply not all switches equal. This should be obvious, there’s a reason all vendors have a wide range of options at different prices, but… what happens if you get the wrong kit?

The most obvious metric is the device throughput – how much traffic can be switched or routed by the hardware – you simply want to avoid bottlenecks where possible and that’s straight forward enough. If you have a 10Gb internet connection and plug it into a 1Gb switch port, you now have a 1Gb internet connection. I have seen that done and the ticket raised “why are we not getting the speed we’re paying for”.

What I’m more interested in is table sizes.

Switches have MAC tables and routers have ARP tables (IPv4) and Neighbour tables (IPv6) and, yes, these are often the same physical devices.

The MAC table records the MAC addresses seen on each port and is key to switching traffic efficiently, sending frames only to the appropriate port. The ARP and Neighbour tables match an IP address to a MAC address so incoming packets can be forwarded. If these tables are filled the router may drop traffic for any addresses it doesn’t know.

Put simply the network equipment needs to have sufficiently large table sizes to accommodate the number of network clients. When this is not the case, things stop working and often in unpredictable ways; some clients will work just fine, others won’t work at all. Maybe someone was happily on line at 10am but at 10:11am when they try again, nothing is working.

This is a common problem with cheap gear. The ISP router might let you setup a really large subnet but it may only be able to handle 50 clients at once due to memory limitations. It’s weird to refer to network gear as “prosumer” but there’s plenty of more affordable kit out there that will take the config you want without delivering what you need.

With serious enterprise gear the table sizes are often really big but the devil is in the detail. A switch might have full layer3 support but be weighted towards layer2 tables with a HUGE MAC table supporting hundreds of thousands of MAC addresses but perhaps a much smaller layer3 table of only 16,000 or so. Because you usually want one or the other the switch may allow you to configure which table is larger.

Here’s a little tale from personal experience. Consider a university campus wireless network which could see over 20,000 clients. As part of a general upgrade the routing was moved from an aging Procurve 5400 switch to a Comware 5900. Everything was great until, a couple of weeks later, term started and the client numbers climbed past 16,000 at which point the calls started coming in about the Wi-Fi not working.

That number was very suspicious and very quickly the recent changes were looked at. We found the new 5900 switch had a ARP table limit of 16,384 vs the MAC table limit of 131,072. This was a switch optimsed for layer2 switching in the data centre, lots of VMs and therefore lots of MAC addresses. But the small ARP table made it unsuitable for our use case. We didn’t know this until it became painfully obvious.

Finally a word of warning on IPv6. Table sizes need to be much larger for the same number of clients. With IPv4 each client will have one address. With IPv6 clients will have multiple addresses. IPv6 addresses also take up more space so the same physical memory will handle far fewer IPv6 table entries vs IPv4.

Understanding the capacity of your equipment is just as important as the functionality.