Fork me on GitHub

Today

I did some more research on SDN today. I learned that my suspicion was correct in that if a port fails the switch would have to request a new flow from the controller. This would significantly increase the failover time, I read some examples of 300ms. Obviously this is too slow for protection applications and is even slower than RSTP.

However, version 1.1 of the OpenFlow standard introduced the idea of Groups. According to this article, OpenFlow Fast Failover groups are specifically meant to address this issue. The article says the failover times are not defined by the standard but are meant to be faster than having to request a new flow from the controller.

I believe this is how SEL is achieving 100μS failover times with their SDN switch product.

Comments