Standard for bandwidth utilization over WAN circuit

Is there any type of industry standard threshold for bandwidth utilization over a WAN circuit? In other words, and regardless of my baseline, when should we consider adding more WAN capacity? Our current methods use the following:

Average daily performance during normal business hours (8-6 local time)
<38% = stable
38-42% = borderline
42-47% = concern
>47% = definite upgrade
Is this valid?

Thanks for you time.
This is a great question. It looks simple but only on the surface.

The first question should really be: Is there any type of industry standard for the definition and subsequent measurement of utilization?

We all have an intuitive notion of what utilization means - maybe it starts by picturing the WAN circuit as a pipe of a certain diameter and then imagining that it is partly filled with something we call traffic. Then we try to relate that notion to facts that we actually have access to - maybe packet counts on a critical router over a certain period of time.

But the fact is traffic isn't some steady-state phenomenon. It fluctuates. A lot. And sometimes very very quickly. Some studies (see link below) have described the appearance of aggregate backbone flows as "bursty." The observation is that a link can be fully utilized in one moment, then completely empty the next. And on a very wide range of time scales, from milliseconds to minutes, and in some cases to years.

So when you measure "43% utilized," what are you really measuring?

You can probably come up with a reasonable definition for utilization. But it is unlikely to be the same definition as the next guy. You are talking apples and oranges, often depending on what equipment you have, where you get your numbers, and what kind of flows you have over the link in question, and what period of time you are sampling over.

So I am curious about what your "current methods" are. I might be able answer more coherently knowing what they are, and what metrics for measuring you employ. And most importantly, the kinds of applications depend on the networks and thus define those metrics.

My response becomes kind of arbitrary otherwise. Good clean links carrying continuous, largely one-way flows like data transfers can be quite solid up to 60-80% utilization (using some protocols). But even slight packet loss over sufficiently long WAN circuits will seriously degrade TCP transfer capacities regardless of utilization. Large variation in RTT due to congestion even without any loss will cripple interactive applications like VoIP.

It all depends on lots of factors and what is important to you. So you tell me.

Dig Deeper on Network infrastructure

Unified Communications
Mobile Computing
Data Center