Three thoughts to consider on how they are taking practical steps forward
Using inexpensive, “bare metal” switch hardware is the first. By unbundling networking hardware and software, the rate of evolution for them increases. They are no longer subject to an engineering ‘least common denominator’ between networking hardware and software innovation.
They are using software defined networking principles -- centralized control for extreme automation. While there is a spectrum of implementations here, the details matter less than the simple observation that the hyperscale data center operators have turned research in to production vastly faster than incumbent vendors.
And lastly are core and pod data center network designs, a break from the traditional “core-aggregation-edge” data center networking mentality. The design specifics don’t matter. What matters is that the hyperscale data center architects are building networks that are emerging optimized for evolvability, optimized for changes to networking engineering itself, and are able to manage huge installed plants yet push the price/performance/features curve.
Network engineering has changed very little since the late 90’s when we started Juniper. It is time to think about re-engineering network engineering. Abstraction and modularity are the solutions to everything, and just as these concepts are changing the rest of computing, they are changing networking. We funded a few companies early on, like Big Switch, that combine these 3 observations to bring hyperscale networking to a broader audience. I am excited to say we are seeing the realization of this vision.
-- Vinod Khosla
This guest blog post is part of the Big Switch Networks' Summer Launch.