Improving energy efficiency of Ethernet switching with modular Cuckoo hashing

Abstract

In recent years, energy efficiency has become an important design goal for routers, switches and other networking equipment. Traditionally, the energy consumption of these devices has been almost constant regardless of the traffic load and the device configuration. In computer networks, traffic loads typically present large variations and in many cases the links operate most of the time at low/medium loads. Therefore significant energy savings can be obtained by making energy consumption proportional to the traffic load. The same applies to the switch configuration that determines the processing done to incoming packets. In most cases, the settings are far from the worst case. This means again that by adapting the power consumption to the configuration settings substantial savings can be achieved. The importance of configuration and packet processing is poised to grow with the adoption of Software Defined Networking. Therefore improving its energy efficiency is important. In this paper, a step is made in this direction by introducing modular Cuckoo hashing an implementation of Cuckoo hashing that adapts its energy consumption as a function of the number of entries. Cuckoo hashing is commonly used in networking equipment to implement packet matching against a set of stored entries. The proposed scheme varies the number of hash tables used as a function of the number of entries thus enabling to power down memories and save energy. Modular Cuckoo hashing is evaluated by simulation and the estimated benefits in terms of power consumption are reported. The results show how modular Cuckoo hashing can make the energy consumption adapt to the number of entries. The practical benefits are also illustrated with a case study of data center switches. This is just an example but makes a strong case to develop more general energy efficient packet processing functions that adapt their power consumption to the configuration settings.