FlowShadow: Keeping update consistency in software-based OpenFlow switches

Abstract

The fast path, as the cache of exact-match rules in the slow path, is applied in software-based OpenFlow switches to improve the forwarding performance. A microflow in the fast path is the specification of its corresponding rules in the slow path, i.e., every field is explicit in a microflow. A rule can generate multiple microflows in the fast path, and a microflow can be generated from multiple rules since there are multiple flow tables in an OpenFlow switch. Due to the many-to-many mapping relationship between the microflows and the rules, the update consistency between the slow path and the fast path becomes a big challenge in software switches, e.g., Open vSwitch (OVS). In this paper, we propose a cache-based scheme (named FlowShadow) to achieve high update performance while keeping update consistency in OVS. In order to examine the reliability, validity, utility and scalability of FlowShadow, we implement FlowShadow on the OVS and conduct numerous experiments with different settings to measure the performance of FlowShadow. The experimental results demonstrate that FlowShadow achieves a lookup speed of 75 million packets per second on a commodity PC under the real backbone traces; the system with FlowShadow speeds up 3.4× times of the original OVS; and FlowShadow also shows high update performance and good scalability at different update speeds and with different numbers of flow tables.