Collect of network telemetry metrics in programmable switches with multiple queues.

• Grant number: 22/13544-8
• Support Opportunities: Scholarships in Brazil - Scientific Initiation
• Start date: December 01, 2022
• End date: May 31, 2024
• Field of knowledge: Physical Sciences and Mathematics - Computer Science - Computer Systems
• Principal Investigator: Fabio Luciano Verdi
• Grantee: Mateus de Nuzzi Bragatto
• Host Institution: Centro de Ciências em Gestão e Tecnologia (CCGT). Universidade Federal de São Carlos (UFSCAR). Campus de Sorocaba. Sorocaba , SP, Brazil
• Company: Universidade Estadual de Campinas (UNICAMP). Faculdade de Engenharia Elétrica e de Computação (FEEC)
• Associated research grant: 21/00199-8 - SMART NEtworks and ServiceS for 2030 (SMARTNESS), AP.PCPE

Abstract

Monitoring together with management are the foundation for any mechanism capable of orchestrating flows on the network. In the context of computer networks, over the years strategies have been developed aimed at maintaining the quality of the services that run over these networks. Typically, a Service Level Agreement (SLA) is made between the customer and the service provider, in which acceptable limits for providing a service are established. Violations in the SLA happen when these limits are not respected, which lead to punishments provided for in the contracts and which can lead to financial losses for the contractor. For this reason, dealing with the quality of the services that run over the networks is a recurring topic. With the emergence of modern networks and programmability in the data plane, a new range of monitoring strategies and management actions emerged, presenting new horizons for network operators. One of them is the INT (In-band Network Telemetry), in which telemetry instructions in the data plane allow the collection of new metrics related to the buffers of the network interfaces, at a level of granularity never reached before. A collection at this level of granularity generates an abundance of monitoring data, which can be useful for Machine Learning algorithms as they depend on large amounts of data. Current works consider the monitoring of network interfaces assuming that they have only one packet queue. However, the latest programmable devices have between 8 and 128 logical queues, each of which can be allocated to flows with different quality of service. In this sense, this research project presents a proposal to design and implement a queue occupancy monitoring solution in programmable networks using INT. To carry out the project, we will implement the solution (or solutions) in a virtual environment using initially BMv2 switches and, later on Intel/Tofino equipment for evaluation in programmable hardware.