It's not easy to interpret whether bandwidth usage is "good, oversized or under-utilized," based solely on a bandwidth utilization graph, and the best benchmark for measuring if a network link is over-utilized is to measure the impact of the resulting network bottleneck on the critical business applications and users using that link. The following are examples where high and low bandwidth utilization does signal the types of link upgrades...
or downgrades you would expect.
Business versus non-business traffic
Most companies don't want to invest in upgrading their WAN/Internet bandwidth for a saturated link that is filled up by employees who are Web surfing and listening to Internet radio. Now, if a link is saturated with critical SAP traffic, and it is impacting employee productivity, then that is a different story.
Quality of Service (QoS)
While a link may have high utilization, it may not be a problem because priority traffic is still routing properly, and therefore, users are experiencing very little delays. Under these conditions, you could be running at extremely high bandwidths, yet not need to upgrade any of your links. However, if the available bandwidth for priority traffic is starting to reach the capacity of the link, then you may have to start thinking about upgrading your link.
Although a link may show low bandwidth utilization, packets may be getting lost across the WAN or Internet. Typically, where there is packet loss, users experience slower application performance because data needs to be re-sent (retransmissions), and the user must wait for this re-sent data. In this case, there is no need to upgrade the link speed. However, you do need to call your service provider and ask for a fix to the packet loss problem.
The following information is typically furnished by an ideal performance management solution:
- Real-time and historical trending information.
- Auto-discovery of "Who" and "What" is using the network. Typically, this data consists of the top IP addresses, IP conversations (source and destination IP address), and applications being used.
- End-user performance measurements, including application response time, network latency, and delay, due to retransmissions.
This question was asked at Ask the Experts on SearchNetworking.com.