Streaming telemetry is all the rage, and for good reason–it gives you immediate data that can inform your monitoring and analysis for performance, operations, and more.
However, today’s show dives into an architecture that forgoes streaming telemetry. Our guests looked at it, considered it, and decided it wasn’t a fit. We’ll talk about why they went in a different direction, the problems they’re trying to solve, and how it’s working out.
Our guests are Kevin Landreth, Director, Service Reliability; and Carl Montanari, Network Reliability Engineer, both at Packet Fabric.
- What we mean by streaming telemetry
- Pros and cons of streaming telemetry
- Why polling still has value (based on your requirements)
- The need for context
- How they get visibility and granularity
- Data volume and scale
- Polling isn’t *just* about time-series type data — we poll ALL kinds of data and use that data to inform operations across our network — context!!!
- Streaming tends to be much less resource intensive, but is not necessarily a panacea — CONTEXT!!!
- Easier to scale *out* polling — scale it out closer to actual devices (geographically) as well as just scaling out worker nodes (our middle-ware tier for example) in a relatively easy fashion
Join ThousandEyes on July 16th for the State of the Internet, where ThousandEyes will unveil its latest research on Internet Performance — a measurement-based study of the availability and performance of the Internet and key app delivery networks, including public clouds, CDNs, and DNS providers. Hear from experts and leaders in Internet and application delivery including Geoff Huston from APNIC and David Belson of the Internet Society. Sign up at thousandeyes.com/state-of-the-internet-2020 to save your spot, or register for the on-demand recording to watch at a later date.
@carlrmontanari – Carl Montanari on Twitter
Day Two Cloud 053: Effectively Monitoring Cloud-Native Applications – Packet Pushers