First, let me quote a from a job candidate.
He has subsequently accepted our offer but it is a real typical one: "Can a centralized openflow based model for running complete complex network control plane is fully scalable, practical, deployable, interoperable, etc. when compared to the current traditional way of establishing packet switched network which are highly distributed for the longest time."
The answer is, "Oh Hell No"!
Secondly, if replacing the existing infrastructure isn't what SDN/OpenFlow about, then where is the money for SDN/OpenFlow?
IMHO, the golden goose is about complementing existing infrastructure with unprecedented flexibility, agility, visibility, resource fungibility in a way that is multi-vendor, multi-skillset, and compatible with today's toolchains and workflow when needed.
For instance, in any data center, if you ask a server admin, "your life changed in the last 10 years?", the answer is obviously "Oh Yeah, instead of dealing with physical machines and physical racks, I'm mostly dealing with virtual machines with mouse clicks"; if you ask the same question to a network admin, the answer will be "Not at all, I'm still dealing with the same VLAN and ACL CLIs that I had 10 years ago, it works but I simply cannot keep up with the mouse clicks and all the fancy ideas from the server guys".
The fundamental issue here is the disparity between "machine speed" vs. "human speed", network admins are desperate about a technology that gives them the "machine speed" otherwise they are in the way of the business.
Lastly, are those golden goose use cases implementable without forklifting?
Absolutely possible at least. In some cases, OpenFlow is a feature of the switch but not a wholesale deal. In other cases, OpenFlow is augmenting the networking, not replacing the network.
All in all, SDN/OpenFlow is about complementing not forklifting existing network infrastructure, it is just like VMware server virtualization technology complemented previous generation server infrastructure.
1 comment:
Howie,
Thanks your for those perspectives.
Yes for sure, the existing infrastructure seems indispensible for a while .
However, do you think that the current distributed path computation models [OSPF, BGP, ISIS, etc] would start becoming challenged as people realize that very same problems can be solved in cluster computing environments?
Added to the fact that they were solved in a manner which was prevalent (low power CPU, low memory footprint) during the times they were invented, would they not start to become obsolete?
Post a Comment