By: Frank Brockners

Frank Brockners — Mon, 16 Jul 2012 14:23:40 +0000

Thanks Brent. You touch on an important point – “we are chartered towards more layering and abstraction”: I agree, network abstractions are unlikely to be completely generic, but will factor in domain and task specific requirements and constraints. Just consider a “network topology” abstraction: The requirements for scale, performance, availability, and accuracy differ quite a bit for applications such as network visualization, service placement, layer-3 style forwarding control (with TTL), or layer-2 style forwarding control (without TTL).

0 likes

By: Brent Salisbury

Brent Salisbury — Sun, 15 Jul 2012 19:54:50 +0000

Thanks for sharing Frank,

That sounds like a central control point to execute PBR based on CPE policy into an IGP via onePK.

I agree a hybrid approach is the obvious first step in the evolution. I happen to believe in centralization still, there are varying degrees to what we are centralizing. A tree approach Islands are a popular thought today for migration paths. Creating subdomain structure just as we do for the same reasons today would likely fit the same construct. Peeling traffic to a different pipeline looks pretty elegant also for the hybrid approach.

A more rich hierarchy and layered approach to state synchronization is likely in order. In your OSPF example I believe there could be consolidation into a controller sub-domain with state synchronization vertically of horizontally from there. We seem charted towards more layering and abstraction. Synchronization is a big challenge good thing we have Google doing what they are on that front

Good read.
-Brent

0 likes

Comments on: Distributed? Centralized? Both?

By: Frank Brockners

By: Brent Salisbury