Performance Analysis And Optimization Of The Sleeping Strategies For Energy-Efficient Ethernet

Due to the increase of the link rate and the network growth,the huge number of Ether-net devices have imposed a heavy energy burden on the communication networks.To relieve the power consumption of Ethernet devices,the IEEE published the Energy-Efficient Ethernet(EEE)standards,which define low-power sleep modes for the Ethernet interfaces with different data rates.Accordingly,several sleeping strategies were proposed,aiming to maximize the en-ergy saving by employing the sleep modes.However,the transitions between active and sleep modes,i.e.,the Sleep and Wakeup operations,introduce energy and delay overheads.There-fore,the key issue in sleeping strategy design is how to properly select the system parameters so that the mode transition overheads can be reduced.To solve this issue,this dissertation first models the behavior of the sleeping strategies for the 10G EEE and the 40G/100G EEE.Based on the analytical results,it then summarizes the threshold selection rules for each of these two strategies,which provides a guidance for the application of EEE in practice.Specifically,the main contents are as follows.Based on the characteristic of the sleeping strategies,we first develop a model for the M/G/1queue with vacation times governed by the arrival process.When implementing the sleeping strategies,the EEE interface first enters the sleep mode each time the buffer becomes empty.As soon as the amount of accumulated frames reaches a certain number or the waiting time of the first frame reaches a certain time,the interface alters its state,e.g.,switches to another sleep mode or wakes up.Essentially,the working process of the sleeping strategies can be modelled as an M/G/1 queue with vacation times that are governed by the arrival process.However,the existing methods can only tackle the system when the vacation time is independent of the arrival process.To delineate the behavior of an EEE interface,we develop a new approach.The key idea is to derive the mean delay performance by starting with the distribution of the number of arrivals during the vacation time.Furthermore,we show that the classical P-K formula of the mean delay is only a special case when the vacation time is independent of the arrival process.Thus,our analytical method is a generalized extension of the classical model.Based on the above model,this dissertation studies the sleeping strategy for 10G EEE.The10G EEE interface employs a single sleep mode,i.e.the Low Power Idle(LPI)mode,to save energy.The burst transmission(BTR)strategy is mostly used to govern the sleep mode of an interface.Each time the buffer becomes empty,the interface with the BTR strategy enters the LPI mode through a Sleep operation.Then,it uses two thresholds,counter thresholds N and timer threshold?,to control the duration of the LPI mode.We first derive the distribution of the number of arrivals during a vacation time based on an event tree of the BTR strategy,from which,we then obtain the power efficiency and the mean delay.Our analysis demonstrates that the BTR strategy has the following two features.First,thresholds N and?are compensating each other.Thresholds?ensures the frame delay is bounded when the traffic load is low,while thresholds N ensures the queue length at the end of vacation times is bounded when the traffic load is high.Second,with the increase of N and?,the power efficiency increases and finally converges to a constant while the mean delay linearly grows.These results,in turn,provide a set of threshold selection rules.Our simulation results show that the BTR strategy with our proposed rules can adapt to the fluctuation of the bursty input traffic and thus performs well.This dissertation further studies the two-stage strategy for 40G/100G EEE.The 40G/100G EEE employs two sleep modes to save energy:fast wake(FW)and deep sleep(DS).With thresh-olds N_fand T_FW,the two-stage strategy selects the sleep mode according to the traffic rate.As a result,the working cycle of the interface switches between two types:deep-sleep(DS)cycles with DS mode and light-sleep(LS)cycles without DS mode.Motivated by this observation,we propose an analytical model based on the concept of conditional sleep-mode,which treats the performance of the two-stage strategy as the weighted average of that of the DS cycle and the LS cycle.Finally,we obtain the closed-form expressions of the power efficiency and the mean delay.The analysis demonstrates that if setting thresholds N_fand T_FWaccording to the aver-age traffic rate,the interface will almost enter DS cycle and thus achieve high power efficiency when the traffic rate is low;while almost enter LS cycle and thus suppress the queue length when the traffic rate is high.Based on these results,we propose four threshold selection rules for the two-stage strategy.The simulations show that the two-stage strategy with these rules can select suitable sleep modes according to the instantaneous traffic rate,and thus outperform other strategies under bursty input traffic.In summary,the contribution of this dissertation are twofold.First,we develop an analytical model for the M/G/1 queue with vacation times governed by the arrival process.This model not only provides a theoretical method to analyze the sleeping strategies for the EEE,but also generalizes the classical M/G/1 queue with vacations model.Second,we derive the closed-form solutions for the performances of the BTR strategy in 10G EEE and the two-stage strategy in40G/100G EEE.Based on the analytical results,we propose the rules to select proper thresholds for each strategy,which can provide guidance for the application of EEE in practice.