As data centres grow to hyper-scale, the limitations of traditional Layer 2 networking architectures are becoming apparent. Issues such as ToR TCAM explosions, large L2 broadcast domains and slow STP convergence can limit scalability and performance. One approach to resolve these issues, which has been successfully deployed in hyper-scale data centres, is the use of fully Layer 3 networks using Layer 3 routing protocols such as BGP or OSPF. BGP, in particular, has proved it's scalability as it is the main routing protocol on the internet and there is a growing trend to use it internally to route traffic in L3-routed data centres.
This paradigm shift in data centre design presents challenges in some of our products such as OSP and OCP which have been designed with the assumption that L2-switched networks are the predominant networking architecture in the data centre. In this presentation, we will discuss some of the key uses cases for which we expect Layer 3 routed networks will be beneficial. We will present some of the architectural and software components that will need to be deployed to enable BGP across our products and ongoing work in open source communities to enable them. Finally, we will present what the future may hold and some challenges that will be presented across our products in order to further enable these use cases (workloads that rely on L2 and hardware offload).
Daniel has worked for the past years on OpenStack networking contributing to Neutron and its integration with OVN.His past professional experience includes embedded systems, electronics and cryptography.