Modeling the impact of pre-eviction time on utilization of SDN flow
rules for adaptive timeout policies
Abstract
Software-Defined Networking paradigm allows the control plane to
centrally supervise the data planes for improving network manageability.
It parallely enforces customizability and programmability at the
switches. TCAM, an expensive, and compact memory is used in the switches
to store flow rules, procured from the controller. Because of the memory
constraints, the unused flow rules should be expedited in timely manner
to reduce flow table occupancy and miss ratio. Simple Hard and idle
timeout fields of the flow rules trigger automatic eviction. The
objective of this work is model the performance of the flow rules which
use dynamic timeouts in SDN flow tables. Literature reveals that
eviction policies; Least Recently Used (LRU) and Adaptive Timeouts were
proposed for enhancing the performance. Many of the works are targeted
towards measuring the performance outcomes through simulations. In this
paper, We proposed a novel idea of measuring the utilization of flow
rules for adaptive timeouts through its pre-eviction time that captures
the idle time since its last usage to eviction.Through mathematical
modeling, we have studied the impact of pre-eviction time on utilization
for adaptive hard and idle timeout policies. Our theoretical analysis
illustrates that Adaptive Hard timeout policy experiences less
pre-eviction time than Adaptive Idle timeout policy and yields better
utilization. We have also done brief experimental simulation to
demonstrate our findings practically.