An Efficiency Pipeline Processing Approach for OpenFlow Switch

Abstract

One primary component of OpenFlow switches is a pipeline of flow tables. Flows are directed through the pipeline by looking up a matched rule in each table. Sequentially, Packet matching starts at table 0 and continues to additional tables of the pipeline if necessary. The lookup stops when one flow entry is matched or the end of pipeline is reached. Though effective, this manner of processing is not efficient when popular rules are installed in back tables. If frequently matched rules appear earlier in the pipeline, the procedure of lookup and comparison can be improved. Unfortunately, a simple sorting algorithm is not feasible. In this paper, we formalize the problem of reducing lookup times, which is proven to be NP-hard. A heuristic approach, MILE(migrating flow rules), is proposed to minimize the average number of lookups. Experimental results show that MILE is able to reduce table lookups by 50%.