In the recent discussion about potential replacements for a Cisco CSS deployment, the topic of bandwidth requirements came up. How much is enough, and how much is too much?
Ask any network admin (or home Internet user) and they’ll tell you there’s no such thing as too much bandwidth.
Shawn Nunley brought up a great point on bandwidth expectations. While I don’t recall anyone ever saying we’d never need Gigabit interfaces for websites, humanity does have quite a knack for easily consuming more and more bandwidth.
Still, bandwidth is a finite resource, and while capacity is increasing over time, the Internet is unlikely to double in bandwidth every week. I’m lucky if my cable modem provider ups the download speed once every two years (I’m still at 5 Mbit. What are we, cavemen?)
So when doing capacity planning, you have a couple of semi-predictable factors in terms of bandwidth usage. You have your business growth, and you have the ability of the Internet to push traffic.  Both (hopefully) will increase over time, and in not too long this constant (if unsteady) growth should mean we all get Gigabit Ethernet to the home, and if your infrastructure doesn’t have a 100 Gigabit pipe to the Internet, you might as well set up your business in a van down by the river.
That day isn’t today. But today, or some day soon, you’ll need to buy some network gear. So do you drop every dollar you have on a massive piece of hardware that will serve your networking needs for the next 10 years? Probably not.  Likely, you’ll need to temper your financial situation with your growth potential and somehow find a balance.
One method I use to help achieve this balance is to know how long I’ll have that piece of hardware in service. Modern technology has a pretty short shelf life, and it helps to keep in mind every time you purchase something, that you’ll likely have to purchase something down the road to replace it. And the circle of cash continues.
The good news about this continual renewal is that you don’t have to purchase a product now and need it to be able to push 100 times the capacity you currently use.
The typical figure I use for hardware is 3 to 4 years. Networking gear (without a hard drive) can last longer, and some server gear might not last that long (as computational requirements increase and slower systems give up their rackspace for speedier replacements), but 3-4 years is a pretty good rule of thumb.
So when I spec out hardware, I use 4 years as the date to replace it, and try to reasonably forecast capacity utilization during that time. If you’re pushing 10 megs now, doing 8 times as much traffic still means a 100 Mbit uplink interface on a load balancer would be sufficient. If you’re pushing 50 Mbit today, however, 100 Mbit probably isn’t going to cut it.
In the next year or so, it will probably get to the point where it doesn’t make sense to get a load balancer that doesn’t have a Gigabit Interface, even for the low-end. Today, however, there are still cases in the SMB where 100 Mbit interfaces should be enough for a couple of years.
And if you experience some massive that dramatically reduces the shelf life of your gear, that’s usually good news, and additional revenue comes in to help pay for the new gear. Again, it’s a balance on future use versus current financial realities.
If course there are always unknowns (known unknowns, or is it unknown unknowns). Perhaps Web 3.0 comes out in two years and HTTP is replaced by IBWP (Insanely Bandwidth-Wasting Protocol) and even your small setups require multi-Gigabit connections. But that’s pretty unlikely.
