TFI2014 Keynote - Russ White
Download as PPTX, PDF
0 likes251 views
This document discusses the future of software-defined networking and some challenges that need to be addressed. It notes that there is no single solution for complexity, and that failure domains and the separation of policy from reachability need to be rethought. The document predicts that while SDNs will have an impact by moving some complexity to centralized control planes and reshaping failure domains, the changes will not be as clean or in the way expected. Personal predictions are that coding skills will become more common while the command line and clean implementations decrease in importance.
TFI2014 Keynote - Russ White
- 2. You are here
- 3. SDN Falls Is this the future?
- 6. You are here
- 7. Where do we go from here?
- 8. THERE IS NO SILVER BULLET FOR COMPLEXITY C1/R Impossible!
- 9. WE NEED TO RETHINK FAILURE DOMAINS Wish you were here
- 10. WE NEED TO SEPARATE POLICY FROM REACHABILITY !=
- 11. WE NEED TO RETHINK OUR MODELS
- 12. WERE NOT GOING TO SOLVE DIFFERENTIATION != != != !=
- 13. THE NETWORK STILL ISNT FREE Never underestimate the bandwidth of a station wagon full of tapes hurtling down the highway. Andrew S. Tanenbaum
- 14. WHAT WILL THE INTERNET LOOK LIKE? Messy, as always SDNs will be overhyped and overplayed Back in the real world, SDNs will have an impact Some complexity will be moved from distributed control planes into centralized control planes Failure domains will be reshaped Policy will be separated more cleanly from reachability But none of these are going to be Done in the way we expect As clean as we might expect
- 15. PERSONAL SCALE THOUGHTS The command line will become less important Theory and principle will become more important Coding skills will become the norm
Editor's Notes
- #9: Solving Complexity is simply impossible SDNs arent going to solve complexity We can move complexity around, though we need to be aware of whether were doing so in an efficient way or not
- #10: Aggregation? Layer 2/Layer 3? We did those things for a reason Moving all our control plane into a single god box that relies on software skills might not be the best answer here Failure that spreads across a single domain is still a train wreck
- #11: We have always put policy into our forwarding tables through communities, metrics, etc. The muddle between these two is making our lives miserable Were trying to build our taxonomy on hard (corner) cases the platypus is not a good place to start when defining the difference between mammals, reptiles, and birds Many of the complexity problems we face today are because were dispersing policy in a thousand different ways, using a thousand different tools; theres no consistency Forwarding needs to be a combination of policy and reachability, but routing doesnt necessarily
- #12: Fast == Control Plane, Slow == Management just isnt going to cut the mustard any longer The seven layer model is (pretty much) dead obsolete We need to think about iterative models, hybrid models lets go back and rethink how we describe networks
- #13: Vendors are not going to lay down and just go with the white box To stay in business, you differentiate your skills, network, etc. Even car vendors differentiate! Vendors want to sell stuff, so theyre going to differentiate There is a rebalancing thats going to happen between commodity and custom, but were still going to have different features, etc. And we wouldnt want it any other way otherwise, you can kiss innovation goodbye
- #14: Networks are a sop to human impatience This isnt about to change any time soon You cant make money shipping bits the bits have to mean something You cant make money shipping content without a container, either Cloud is, to some degree, predicated on the idea that the network is free