Where they can, data networking equipment vendors like to arrange their proprietary products into vertically integrated stacks, with complex functions often "baked" into the hardware. Furthermore, a sometimes tortuous standardisation process makes it hard to implement changes and raises the barriers of entry to new equipment …
TCP/IP time warp
What has TCP/IP got to do with switching? Maybe you're thinking of ethernet or ATM or MPLS?
We've had software defined *routing* for donkey's years. How much effect has that had on the upper layers of a networking stack?
The world of networking seems to have failed to implement ancient standards like multicast over any sort of scale, let alone merely old standards like IPv6 to everyone's satisfaction. What on earth makes you think that something new and shiny is going to come in and sweep the world?
You missed the point
OpenFlow directly updates the FIB in the router. Routing protocols only update the RIB which is subsequently updated to the FIB. This is a significant difference.
Instead of letting an autonomous system propagate data to it's neighbour, a central controller will have a complete view of the network - make some sort of programmatic decision and then download some configuration to the FIB on the switch/router.
In the same way that VMware allowed the effective management of hundreds of windows servers, OpenFlow hopes to provide effective management of hundreds or thousands of network devices as a coherent whole.
Which is much more advanced than a MS server can do today.
Stanford's re-invented spanning tree.
this has the possibility of creating a paradigm shift in networking technology especially in the data centre. There is a really good discussion about it in episode 40 of the (excellent) packetpushers podcast.
Re: actually →
Nice to see at least one person understood the concept.
If all network devices can talk OpenFlow and OpenFlow provides all the features that I require, my network kit becomes vendor agnostic and the premiums that some vendors charge can be greatly reduced or an alternative vendor selected.
Just like the x86 server world where HP/Dell/IBM/other vendor server can be used to perform function X, but one maybe more suitable on price/performance/non-standard feature...
There is nothing new under the sun: Ipsilon, eventually bought by Nokia, did this in about 1997
I was doing this with IBM's internal network in 1985 ...
Dynamic bandwidth allocation, out of bandwidth signaling, and spanning tree algorithms are nothing new.
Nor is using them together on the same network.