“Smart telemetry data processing on network elements – right at the source” par Frank Brockners (Technology and Architecture Office, Cisco)
“Network elements create an enormous amount of telemetry data today. The amount of telemetry data created is much greater than what can be exported by the network element.
So far, network devices use simple techniques like random sampling to reduce the amount of available telemetry information so that such data can be exported to collectors or analytics subsystems. A common train of these techniques is that “interesting” data which would help identify network issues is often left out and not analyzed: “You can only digest what is presented to you”. This talk discusses and presents results of algorithmic approaches deployed on the network device itself to process the telemetry data more intelligently, yet in a very light-weight and fast manner. With that, “interesting” data can be filtered out by the network element, so that only that part of the telemetry data, which is likely to be of use for further processing by analytics subsystems, is exported.”