In band Network Telemetry (INT) allows for a high accurate monitoring of the network status, at the cost of an increased overhead, due to the insertion of telemetry information in the packets carrying the user data. In this paper we present a novel approach to reduce the overhead due to the realization of network monitoring solutions based on the use of INT. In particular we introduce the concept of spatial sampling as a means to reduce the amount of telemetry data to collect. The idea is to exploit the spatial correlations arising among the different INT data and imposed by the network topology, to reduce the amount of telemetry values to collect and reconstruct the unseen ones. We further introduce SPAN, a sampling and recovering strategy that works according to the spatial sampling paradigm. The integration of SPAN in a programmable network is discussed and an extensive performance evaluation is performed. The results show that it is possible to reduce the overhead due to INT data collection of 65%, while achieving a high accurate reconstruction of the network status.
Reducing the in band network telemetry overhead through the spatial sampling. Theory and experimental results
Sardellitti S.;
2024-01-01
Abstract
In band Network Telemetry (INT) allows for a high accurate monitoring of the network status, at the cost of an increased overhead, due to the insertion of telemetry information in the packets carrying the user data. In this paper we present a novel approach to reduce the overhead due to the realization of network monitoring solutions based on the use of INT. In particular we introduce the concept of spatial sampling as a means to reduce the amount of telemetry data to collect. The idea is to exploit the spatial correlations arising among the different INT data and imposed by the network topology, to reduce the amount of telemetry values to collect and reconstruct the unseen ones. We further introduce SPAN, a sampling and recovering strategy that works according to the spatial sampling paradigm. The integration of SPAN in a programmable network is discussed and an extensive performance evaluation is performed. The results show that it is possible to reduce the overhead due to INT data collection of 65%, while achieving a high accurate reconstruction of the network status.I documenti in IRIS sono protetti da copyright e tutti i diritti sono riservati, salvo diversa indicazione.