The core principle of model-driven telemetry is super simple: as much usable data out of the network as fast as possible. The implementation is also pretty subtle: a push mechanism, data models, open-source encodings. In fact, it’s all so simple and subtle that some folks look under the hood and say, “so what?” To that, I say…great question!
After years of riding the roller coaster of hype curves (SDN anyone? Cloud? DevOps?), experienced network engineers are rightly skeptical of “the next big thing.” So let’s get at the who should be looking at telemetry and why.
The big Web companies who worked with us to develop telemetry were clear about their first use case: SNMP replacement. As I explained in a previous blog, SNMP simply could not deliver the speed and resolution they needed to monitor their networks. But what if you’re not operating at Web scale? What if you’re fine getting network stats every 15 minutes? What if you like your monitoring tools and they give you all the insights you need?
In that case, the answer’s pretty simple: stick with SNMP. Seriously. I’m not a fan of chasing the shiny new object just because it’s shiny and new. If your monitoring solution works for you, then stick with it. By implementing best practices, most people can probably even squeeze a little more out their current SNMP deployment.
But just to be a little contrarian (this doesn’t come naturally for me, but I’ll try), I would encourage you to take a good hard look at the future. The big winners in the current consumer technology space (see, for example, Farhad Manjoo’s article on The Frightful Five) have been the ones who relentlessly pursue simplicity, speed and automation in their networks. And since you cannot automate what you don’t understand, better network visibility is a non-negotiable prerequisite. To get that kind of visibility, you need more data, faster – in other words, you need telemetry.
Sometimes “more” means more variety. The truth is that SNMP doesn’t always have the data that you need. With hundreds of YANG models available today in IOS XR and more to come, model-driven telemetry exposes a much larger space of operational data than SNMP. But for the initial telemetry use cases, “more” means more of the same old, boring data.
Take the simplest statistics you can imagine: bytes in and bytes out. With some simple mathematical correlations, you can use these statistics to detect dropped traffic, black holes, infiltration attacks, bundle link polarizations, and many other significant network conditions. Some operators already do these calculations with SNMP data. But if you’re only getting data every 15 minutes, you’ve got a built-in limit on what kind of events you can detect and how quickly. By collecting data an order of magnitude faster, you get an order of magnitude better resolution. Boring is really a function of scale. The more data you get, the less boring it becomes.
So, embrace boring! It’s the new awesome. You might not need more boring data today, but you will eventually. Automation is unstoppable and the pace is relentless. Model-driven telemetry will help you keep up.
If you want to learn more about model-driven telemetry, visit us here.