Network engineers spill a lot of ink about redundancy, but the form that redundancy takes depends a lot on circumstances. One of the most important of these circumstances is the size of the network. While there are exceptions galore, in general, the smaller the network, the less redundancy you see.
I’ve built and maintained the network in many small and medium offices. A fairly standard setup for under 100 users is a single router on the edge of the network, and a few switches dotted through the building as needed. On its face, there’s not much room for redundancy there—but there is some. And there are ways to mitigate the lack of redundancy in other areas.
Start with that edge router. Most will offer redundant power supplies, which is important because power supplies are one of the most likely components to fail.
A smaller piece of hardware will have limited options, but usually has enough ports to permit two ISP links (generally a reputable primary link, and a business DSL or similar as backup), and two connections to the LAN.
In a medium or large business with multiple offices, it often makes financial sense to keep a spare router and switch at the central office, ready to ship overnight if one fails at a remote site.
An important failure in redundancy that often goes unnoticed is a lack of out of band (OOB) access. What I’m talking about is some method to connect to the console port on a router, switch, load balancer, or firewall.
Typically this is accomplished through a dedicated broadband, T1, dial-up, or even 4G link. That dedicated link terminates on a console server (or firewall in front of one), which lets engineers then choose which piece of equipment they want to console into.
It’s a lifesaver in a lot of situations, but it usually gets swept aside because it requires dedicated gear, an independent circuit that’s idle 99% of the time, and someone to drive the project. While I’m sure there are plenty of small businesses with OOB access to remote gear, I’ve yet to see one myself.
Medium-sized businesses are usually able to justify and afford a little more advanced gear. In the switching closets, they can use stackable switches—switches that connect together to look like a single machine. Instead of the one or two switches in a small business, they might use two or three that are managed as one.
At that point, the extra cost to add a fourth switch to the stack becomes more reasonable, and the network engineer can fight for funding or quietly overestimate his needs. If a single switch fails, he can walk over and move cables and be up in minutes.
The edge router in a medium-sized businesses usually turns into a redundant pair, each with an ISP link. And it’s perfectly normal at a medium-sized business to see a chassis-based core switch or two, offering the client the ability to deploy redundant redundancy—that is, redundant switches, each with dual supervisors, power supplies, and spare line cards sitting in the chassis (the latter preferably powered off).
OOB connectivity is still lacking in most medium-sized businesses I’ve run across, but it’s less critical. While a misconfiguration or software bug might still disable the network, this is where we see most single points of failure designed and engineered away.
A final distinction between small and medium-sized business redundancy is the in-house expertise. In small businesses you often see a single person who “knows the network.” There might be a few other IT people, and there’s a chance the email administrator has read a Cisco book and knows the password to get in if the network administrator hits the lottery and quits, but there won’t be regimented knowledge sharing.
In medium-sized businesses, this has usually been mitigated. You’ll see teams, or at least an IT manager capable of logging into the router and making simple changes. Wikis or other documentation repositories start appearing, and on-call rotations become reality. In short, a medium-sized business can at least survive the loss of any single networking professional, however uncomfortable it may be.
Building redundancy into a network requires planning and, ideally, expertise. Networks that start small and grow organically can end up labyrinths of physical loops and other pandemonium.
But a little knowledge and planning can make a tiny network scalable for almost no extra money by, for example, ensuring the edge router hardware can accommodate a second ISP link, or a redundant neighbor when the time comes. This, I believe, qualifies as “an ounce of prevention.”