What Are Your Favorite Network Health Metrics?

In one of the early design classes, Cisco Systems laid out a nice series of metrics and thresholds that can help you determine the overall health of your network. Way back then, they were speaking of what we now call the Enterprise network.

Let’s check in on these Enterprise network metrics, think about if we still agree with them, and also give you a chance to share your knowledge with others, and maybe even receive some great advice from your peers. Just share your valuable feedback in the comments below this post.

Here was the list (along with a few additions from the Lammle.com training team):

• Ethernet segments should not feature a sustained utilization of 40% or higher
• All Ethernet segments should be switched – no shared segments (hub-based)
• No WAN links should feature a sustained utilization of 70% or higher
• Response times should be generally less than 100 ms
• LAN response times should generally be 2 ms
• No segments should have more than 1 CRC (cyclic redundancy check) error per million bytes of data
• No segments should have more than 20% multicast or broadcast traffic
• For Ethernet segments, there should be less than .1 percent collisions over 5 minute intervals
• CPU utilization should not exceed 75%
• The number of output queue drops should not exceed 100 in an hour
• The number of input queue drops should not exceed 50 in an hour
• The number of buffer misses should not exceed 25 in an hour
• The number of ignored packets should not exceed 10 in an hour

As you look over the list, you are certainly going to recognize some areas you consider to be “off the mark”. In fact, way back when this list originally was published (about 2010), you could have easily taken great issue with many of them when thinking about your own network then. General recommendations are just that – they are general. You might have requirements due to special applications that skew these numbers dramatically. For example, network device CPU utilization might be well under 60% at all times in your Enterprise infrastructure.

We should also remember that today we further divide the Enterprise network into specialized modules and that not all modules will require the same baseline metrics. Perhaps we have some critical services hosted in a portion of the Enteprise infrastructure. The metric values desired in this location could vary widely from the general recommendations.

What are your thoughts? What are some of the key metrics you and your team use when you are proactively monitoring the performance of your Enterprise network? Let us know in the comments below!