Building a transparent Layer 1 Switch using P4

Sebastian Graf

Playlists: 'denog12' videos starting here / audio

This lightning talk describes an internal research project where we wanted to check whether a P4 capable whitebox switch can be used as a cost-effective transparent alternative to existing Layer 1 Switching technologies.

We operate a multi-vendor testing center in Frankfurt that we use for Proof-of-Concept tests, design validations, software qualifications and internal trainings. The involved physical components are interconnected in standard topologies, however we are frequently facing the challenge of temporary connections and limited port capacity on the components.

We used to solve this by providing interconnect switches as part of the lab infrastructure, but faced that problem that this is not working transparent on layer 2. Although vendors implement tunneling techniques for ugly layer 2 protocols, you always find one protocol where this does not really work. The standard solution would be to go down the path of Layer 1 Switches that either work as patch robots, or using the optical-electrical-optical principle.

After checking budgetary offers for these devices, we were motivated to see if anything else could be built with stuff that we already have in our lab and quickly came to our P4 switches.

This talk gives an insight about the results that we got, including the advantages and disadvantages that we see in this approach and also highlights what kind of ugly frames can be forwarded over this solution.

Download

Embed

Share:

Tags