Using Probabilistic Data Structures for Monitoring of Multi-tenant P4-based Networks

Autor(es)

Martins, Regis E. T. ; Verdi, Fabio L. ; Villaca, Rodolfo ; Garcia, Luis Fernando U. ; IEEE

• Tipo de documento: Artigo Científico
• Fonte: 2018 IEEE SYMPOSIUM ON COMPUTERS AND COMMUNICATIONS (ISCC); v. N/A, p. 4-pg., 2018-01-01.

Resumo

With the evolution of the processing capacity of the network devices and the appearing of novel programmable network hardware, new methods and techniques for traffic monitoring have been proposed. Among these new techniques, the use of sketches has been increased. Sketches are probabilistic data structures capable of summarizing information about the network with two main advantages over flow-based counters: low memory usage and adjustable accuracy. Considering this scenario, this paper proposes the use of probabilistic data structures implemented in P4 (Programming Protocol-Independent Packet Processors) to monitor multi-tenant networks. While it is common to see similar solutions for monitoring non-shared networks, as far as we know this is the first work that deals multi-tenant scenarios. We implemented sketches for each tenant so that an isolated network monitoring can be done. The solution, named BitMatrix, was created taking into account two probabilistic structures: bitmaps and counter-arrays. (AU)

• Processo FAPESP: 15/19766-9 - Monitoramento de recursos físicos e virtuais em ambientes de cloud computing e infraestruturas virtualizadas
• Beneficiário: Fabio Luciano Verdi
• Modalidade de apoio: Auxílio à Pesquisa - Regular