A Policy-Aware Switching Layer for Data Centers

Problem: Managing an enterprise or data center network is difficult because middleboxes must be placed on the path of the traffic they must filter. The authors argue that this is a real problem using statistics about data center downtime. My only question here is just how many middleboxes are really used in a typical data center - it would seem that some services don't use that many and won't benefit so much from this work.

Solution: Allow policies to be specified centrally and programmed onto switches, letting them route traffic through middleboxes in the correct order.

Why is it different from previous work? This seems to be the first paper really looking at this problem from a network architecture point of view rather than trying to hack together a solution using existing network hardware features.

Will it be influential in 10 years? I think this paper makes a good point for higher-level policy control over the network and a smarter switching layer, as well as perhaps packet tagging as a primitive. Other network research (e.g. OpenFlow) tries to achieve similar goals of a programmable switching layer.

Criticism: The need to preserve integrity while switching between policy complicates things and to some extent seems to be a good opportunity to design clever algorithms. Would it be possible to take another approach where you drop or delay packets for a bit but make the switchover very fast?