With our background in network measurements, and specifically flow measurements, we see P4 as a possible alternative or complement to existing measurement technologies. Based on research in previous RoN years, where we showed IPFIX’s capabilities of exporting useful features in (IPv6) network traffic, we found the applicability of such Solutions to be limited and even causing vendor lock-in to a certain degree.

The open nature of P4 and the eco-system P4 is evolving in are not prone to these problems, allowing for wider real-world application. and even more flexibillty To achieve useful results, we will focus on implementation prototypes in P4, on hardware.

As P4 is subject of focus among many RoN partners, we can

leverage implementation experiences from other partners, allowing us to focus on validation and assessment of measurements, This can be expressed by the following possible research questions:

  • How can we realise flow measurements using P4?
  • To which extend can in-band lelemetry be used to enhance or validate P4 flow measurements?
  • How do these flow measurements compare with other technologies, e g. NetFlow/IPFIX, qualitatively and/or quantitatively?

In addition to collaboration within RoN itself, we will leverage other resources and connections. Firstly, there is a strong cross-pollination opportunity with another project we are involved in, where we focus on encryption of flow data. As P4 equipment with cryptographic chips will become available, there is a certain overlap in research areas, enabling us to incorporate knowledge from multiple domains.

 Secondly, we will create an in-house test-bed of multiple servers connected via P4 hardware, enabling performance evaluation and validation of P4 implementations.

Another, smaller angle we would like to take in investigating possibilities of P4 is related to routing. Possible research questions in this area are:

  • To which extend can P4 be used to perform or enhance BGP operations?
  • Can we analyse anycast performance by deploying P4 hardware in strategie places in the network?

Deliverables:

  • Report, possibly in forms of a paper;
  • Presentation of ideas or results at RIPE/NLNOG/NANOG or meetings alike;
  • Open sourced code (P4 and/or code related to measurements, validation) where licensing SDKs allows us to do so
  • No labels