Modelling and real-trace-based evaluation of static and dynamic coalescing for energy efficient ethernet

Abstract

The IEEE Standard 802.3az, namely Energy Efficient Ethernet (EEE), has been recently introduced to reduce the power consumed in LANs. Since then, researchers have proposed various traffic shaping techniques to leverage EEE in order to boost power saving. In particular, packet coalescing is a promising mechanism which can be used on top of EEE to tradeoff power saving and packet delay. In this paper, we analyze the interesting and special case of 1000Base-T EEE links, in which power saving operations are triggered only when links are inactive in both transmission directions. We are the first to provide an analytical model for EEE 1000Base-T which accounts for the bidirectional nature of LAN traffic. Our model allows to compute the power saving achieved by EEE, with and without packet coalescing, by using a few significant traffic descriptors. Furthermore, we use real traffic traces to investigate on the performance of static as well as dynamic coalescing schemes. Our results show that dynamic coalescing does not significantly outperform static coalescing in terms of power save and delay.

An experimental power profile of Energy Efficient Ethernet switches

“…Our model is different from previously proposed analytical models of EEE [32,33,34] in several aspects. Firstly the focus is to model the entire switch compared to an individual PHY as in previous models.…”

Packet Coalescing Strategies for Energy Efficient High-Speed Communications Over Plastic Optical Fibers

“…This was also the case for the initial studies on EEE [10][11][12][13][14] that were subsequently validated with real measurements once devices that supported EEE became available. As for EEE, both synthetic traffic and packet traces are used in the simulations presented in this section.…”

“…To avoid excessively long delays, the maxsize threshold criteria is combined with a max-delay threshold that forces the transition to the active mode and subsequent packet departure as soon as sufficient packets have arrived or the waiting delay of the first packet has reached the max-delay limit, whichever occurs first. Thus, fine tuning packet coalescing strategies comprises a tradeoff between energy savings and network performance metrics measured in terms of packet delay [12][13][14].…”

Energy saving in carrier-grade networks: A survey

“…Since then, several works addressed modeling and performance analysis of EEE based on various traffic parameters and with and without consideration to packet coalescing [133][134][135][136][137][138][139]. In [133] authors provide an evaluation of static and dynamic coalescing for EEE in which buffer size and timeout are fixed (static) and adapted to traffic pattern (dynamic). The results show that static coalescing and dynamic coalescing algorithms achieve similar power saving and delay tradeoffs.…”