Concerto: Cooperative network-wide telemetry with controllable error rate

Abstract

Network-wide telemetry requires real-time analysis of a large amount of traffic. Telemetry systems use stream processors to support various applications, and Protocol Independent Switching Architecture switches to reduce the workload on stream processors. Due to the inefficient use of switch resources, existing systems cannot fully reduce the workload on the stream processors. Unlike the existing systems that treat switches independently when assigning tasks, Concerto lets switches work together. The use of cooperating switches means more resources and a further reduction in the stream processor's workload. Furthermore, Concerto can also adhere to a more stringent error rate requirement. Our evaluation shows that Concerto reduces the stream processor's workload by as much as 19x, and under the same workload on the stream processor, Concerto achieves an error rate of 104x lower than existing systems.