Question: Compare and contrast 5 nines reliability vs webscale reliability.
Instead of building highly reliable systems, we try to build highly resilient systems with cheap, unreliable parts. How do you drive 99% of the cost out of the infrastructure, and compensate in software? Webscale business has to always be up. Everyone wants infrastructure that just works, built for demanding environments.
There are a lot of lessons from the webscale world on how to do those types of systems. [We can] take the best from the telecoms side and the webscale side, and bring those two together.
Question: Best of telco vs best of web scale?
Carrier-grade systems are best at resource management. Take what our radio guys have to do, where your footprint is tens or hundreds of thousands of sites in a country, much more distributed than a datacenter. This is an infrastructure that has to be aware of the device and the application that’s on the other end of it.
In the webscale world, infrastructures are relatively centralized. Distributed knowledge should move into the datacenter world: how do we do distributed scheduling, distributed resource management, how can we be more application-aware, aware of end user requirements? Datacenter knowledge [includes] lifecycle management of components, driving out extra components, unreliable individual components. Bringing those things together is what’s interesting.