I’ve been banging on for awhile about the absolute requirement for visibility into the network. It’s not enough to have a network; we must be able to operate it effectively to offer value and insight back to the business.
Arista Telemetry is both a product and a framework and I’m really positive about this openness. Let me explain by cadging directly from Arista’s briefing slides. (Think of it as a GUI-enhanced blog post.)
Telemetry is not monitoring. Monitoring is based on a pull model such as SNMP/RMON where the software polls for state or data.
Telemetry is a stream of data based on triggers built into the device itself. You configure a network device to “send data when” or “do this if” by sending data to a telemetry receiver. You can stream a wide variety of state information from your box to a telemetry application.
Telemetry is delivered using open standards such as NETCONF/YANG, and I think using gRPC protocols, which are open enough because they are widely used. (Standards bodies are years behind the market right now).
These data sources are wider in scope and more functionally useful than SNMP.
Framework Not Platform
I particularly like the approach that Arista has taken with openness by building a framework that allows for a wider variety of choices.
Many customers will want to buy a fixed solution, with known support models and simpler installation. But more and more customers are deploying their own applications (like Grafana). Or maybe you want CloudVision Telemetry for most things, but just one thing sent to Grafana.
If you want a single vendor, choose hardware/engine and telemetry from Arista, but you aren’t prevented from choosing any others.
Consider how different this approach is to that of other vendors that build “closed open” telemetry platforms that restrict access at different parts of the system.
Telemetry uses a graphical interface. Modern Web applications constantly surprise me with their flexibility and capability.
Available Next Year
This product won’t be available until Q4 2016 (so 2017, then). I think pre-announcements suck, and I’m disappointed that we will have to wait half a year (at least) to see the actual product. While enterprises do need months to schedule products into their plans, the concept of announcing products months or years ahead is a bad one.