Abstract

OpenFlow, as the Software Defined Networking (SDN) primitive, provides a simple forwarding plane abstraction, which heavily relies on the fast memory inside the OpenFlow Switch (OFS). OFS components, e.g. flow table, meter table, counters, have to compete for the limited fast memory resource. As a result, only a few counting functions are defined as mandatory in the OFS specification, although a lot of SDN proposals depend on a detailed states collected by the optional counters in the specification. This fact motivates us to rethink the way to maintain counters in the OFS. We propose a new architecture called CACTI, which only consumes several registers in the fast path and moves the completed counters into the on chip RAM like cache in the slow path processor. Theoretical analysis and experiments on the prototype system demonstrated the efficiency of our architecture: CACTI is capable to achieve the throughput of 29.4-39.7M pps packets per second (pps). No RAM resource is needed any more in the fast path, instead, CACTI consumes only 0.24-0.54\% Look-Up Table and 0.35-0.43\% flip-flops compared with the entire FPGA-based OFS design in the fast path, and the unused CPU cache in the slow path.

Full Text
Published version (Free)

Talk to us

Join us for a 30 min session where you can share your feedback and ask us any queries you have

Schedule a call