|
|
Energy Efficient Ethernet (EEE)
IEEE 802.3az Energy Efficient Ethernet (EEE) was recently approved on September 30, 2010, representing the first standard on energy efficiency in the history of IEEE 802.3 [1]. EEE uses the Low Power Idle (LPI) mode to reduce the energy consumption of an Ethernet link. In the LPI mode, data is transmitted in the active state and the link enters the LPI state when no data is being sent. In the idle state, short refresh signals are periodically sent to keep the link alive and to align the receivers with current link conditions.
EEE Operation
Fig. 1 illustrates the operation of EEE. Transition to the low power mode requires ts seconds. While the device is in sleep mode, it only sends a signal for synchronization during refresh interval tr and stays quiet during the interval tq. When there is data available to send, the device takes tw seconds to wake up and enter the active state again.
Fig. 1. EEE operation.
EEE Overhead
EEE may suffer from a significant overhead stemming from the wake time (tw) and sleep time (ts). For illustration, assume that a 10 Gb/s Ethernet link initially is in sleep mode. Upon receiving an Ethernet frame of size 1500 bytes, the EEE port needs to wake up (requires tw time), transmit the frame (requires tframe time), and return to sleep mode again (requires ts time). For 10 Gb/s Ethernet, transmission of the Ethernet frame takes 1.2 µs and the values for tw and ts in IEEE standard 802.3az are specified as 4.48 µs and 2.88 µs, respectively. Thus, the energy efficiency, defined as the transmission time over active time (transmission time + overhead), is about 14%. The situation is even worse in the case of TCP ACK packets (64 bytes) of duration 0.0512 µs, where the energy efficiency reduces to 0.69%. This clearly indicates the need for improvement. The challenge is to ensure the maximal use of sleep mode, thus minimizing the detrimental impact of the EEE overhead.
Trade-off between energy efficiency and network performance
Recently, packet coalescing has been proposed in [2] to mitigate the EEE overhead, whereby a number of packets are collected for a certain amount of time and transmitted as a burst of back-to-back packets. It improves the energy efficiency in some cases, but increases the latency in the network and degrades QoS in terms of end-to-end delay. Packet coalescing lacks an efficient mechanism for scheduling the transmission times of Ethernet frames; that is, properly selecting the active intervals during which the frames will be transmitted on each port. In packet coalescing, a transmission timer is essentially set for each port; once this timer times out, all packets which have accumulated so far will be burst out at line speed. Indeed, the selection of the timer plays a critical role in trading off the power consumption of the port and the delay experienced by the frames at each port. Hence, scheduling the activation events of the ports will be crucial in reducing power consumption as well as guaranteeing the quality-of-service (QoS) in the network, in terms of end-to-end delay that frames will experience. Here, we present an illustrative example to explain the importance of scheduling in greater details.
Illustrative Example: Scheduling
We consider a simple instance where five frames f1, f2, f3, f4, and f5 of different sizes with various delay thresholds arrive at times 3, 5, 8, 9, and 10 respectively (Table 1). Table 1 presents other parameters used in this illustrative example, such as the size and delay requirement for each frame and transmission time of frames. The transmission time of a frame is computed under the assumption of a link rate of 10 Gb/s. For illustration, we assume that two EEE ports (p1 and p2) are available to transmit these frames.
| Frames
| f1
| f2
| f3
| f4
| f5 |
|
| Arrival (unit time)
| 3
| 5
| 8
| 9
| 10 |
|
| Size (Bytes)
| 1500
| 1500
| 1250
| 1250
| 750 |
|
Req. transmission time (µs.)
assuming 10 Gb/s link
| 1.2
| 1.2
| 1
| 1
| 0.6 |
|
| Delay Req. (unit time)
| 7
| 6
| 6
| 3
| 3 |
| | | | | |
|
Table 1. Parameter's value for given frames.
Fig. 2 demonstrates two different possible scheduling alternatives that can be adopted and are illustrated as follows. Fig. 2(a), referred to as Scheduling 1, assumes that the timer expires at time instant 9. At this time, frames f1, f2, and f3 are in the buffer and will start their transmission through port p1. Thus, port p1 will be in busy state from time 9 to 12.4 while transmitting frames f1, f2, and f3, as depicted in Fig. 2(a)-(ii).
Fig. 2. Alternative schedules for given frames.
When frame f4 arrives at time 9 with delay threshold 3, this frame (f4) must be transmitted at most at time instant 12. Since p1 is in busy state until time 12.4, the scheduler needs to activate port p2 to transmit f4 and satisfy its delay requirement. Now, frame f5 arrives at time 10 and can be transmitted through either port p1 or p2. A possible scenario of transmission of frames f4 and f5 is presented in Fig. 2(a)-(iii). An alternative scenario for scheduler 1 is to transmit frame f4 at time 9, subsequently followed by frame f5 through port p2, as shown in Fig. 2(a)-(iv) to reduce the average delay of those frames. Notice that in both cases scheduling 1 requires the activation of two ports. Fig. 2(b) illustrates an alternative schedule, referred to as Scheduling 2, where port p1 is activated at time instance 6 to transmit frames f1 and f2. When frame f3 arrives at time 8, it is scheduled at time 8.4 on the same port p1. Similarly, frames f4 and f5 are scheduled for transmission on port p1 at time instance 9.4 and 10.4, respectively, as depicted in Fig. 2(b)-(ii). Clearly, this transmission schedule is able to transmit all frames within their deadlines using only one port, while keeping the second port p2 in LPI mode, and thus conserve more energy. It is therefore clear that efficiently deciding the transmission schedules is of utmost importance to conserve energy consumption while meeting the delay requirements of the traffic.
Further reading
- K. Christensen et al.,
“IEEE 802.3az: The Road to Energy Efficient Ethernet,”
IEEE Communications Magazine, vol. 48, no. 11, pp. 50–56, Nov. 2010.
- IEEE 802.3az,
“IEEE Standard for Media Access Control Parameters, Physical Layers, and Management Parameters for Energy-Efficient Ethernet, Amendment 5,”
IEEE Std 802.3az-2010 (Amendment to IEEE Std 802.3-2008), Oct. 2010.
|
|
|
| |