Telemetria preditiva baseada no comportamento de fluxos de rede

Abstract

With the growth in network traffic volume, technologies that ensure security and reliability are explored and reassessed to become more efficient. In-band Network Telemetry (INT) is a technique that reports the network state, providing high-granularity real-time reports. The main issue with INT is the use of additional resources due to the increased traffic volume, known as overhead. Several studies have been conducted investigating this issue, but they share the same limitation, they only analyze and report the past. The objective of the present work is to propose a new way to reduce the overhead of network telemetry. This is achieved by reducing the frequency of telemetry insertion in network flows with predictable behavior. Additionally, the algorithm should report a behavior prediction for known network flows. Through comparisons with other algorithms, it was possible to observe a significant reduction in overhead with minimal loss in the accuracy of the reported information.