تجزیه و تحلیل عملکرد از اترنت کارآمد انرژی در سرورهای ویدئو

Current trends on traffic growth oversee a steady increase of video streaming services, and the subsequent development of the associated infrastructure to allocate and distribute such contents. One of the operational costs associated to this infrastructure is the power bill. Therefore any mechanism used to decrease it, reducing also the carbon footprint associated to it, is welcome. In this work we investigate the suitability of the recently standardized IEEE 802.3az Energy Efficient Ethernet (EEE) for video traffic generated by video-streaming servers. The conclusion of the analysis is positive about the achievable energy savings, due to the inherent features of traffic patterns of video-streaming servers which help reducing the number of transitions between active and low-power modes in EEE.

The Internet growth has also raised several concerns about its power consumption and the need to improve its energy efficiency [1]. It is estimated that the Internet accounts for tens of TWh, an amount that could be significantly reduced if energy efficiency policies were applied. One of the first efforts toward energy savings is the IEEE 802.3az (Energy Efficient Ethernet) standard [2] which reduces the energy consumption of Ethernet devices by introducing a low-power (or sleep) mode. As Ethernet is the dominant technology for wire-line LANs, with more than one billion devices already deployed, Energy Efficient Ethernet (EEE) is expected to enable large energy savings [3]. However, previous studies in simulation [4] and with actual power measurements [5] have shown that the performance of Energy Efficient Ethernet greatly depends on the traffic pattern, and may not be as efficient as originally expected. This is due to the large power-mode transition times; that is, most of the power required for the transmission of an isolated Ethernet frame is spent in activating the link and putting it back to the low-power mode, rather than on actual data transmission. On the contrary, when several Ethernet frames are transmitted back-to-back, such a burst of data shares the sleep-to-active mode transition overhead resulting in larger energy savings [6]. Such bursty traffic pattern has been observed in video-streaming traces, which suggests that the adoption of EEE in video-streaming servers may achieve large power savings. Hence, this article attempts to analyze and quantify the suitability of the Energy Efficient Ethernet standard for the service of real time video streaming in IPTV networks, using the H.264 codec which is becoming the most popular video codec [7]. Indeed, given the high popularity of video streaming services such as YouTube, Hulu or Netflix, Internet video is expected to comprise about half of the total traffic share in the Internet by 2016, according to latest estimates [8].

This work has studied the behavior and performance of Energy Efficiency Ethernet in video-streaming servers, showing good results in terms of energy savings. Essentially, the back-to-back transmission nature of the video-streaming sources results in the transmission of a large portion of data per wake-up and sleep-down cycle. Indeed, the number of power-mode transitions is reduced with respect to other traffic patterns, and the energy overhead caused by switching between the active and low-power sleep mode is minimized. The experiments have further revealed that the use of B frames produce larger transmission bursts than those without B frames, hence achieving larger efficiency values. Furthermore, it has been shown that the H.264/SVC codec not only increases the compression ratio with respect to H.264/AVC but is also more energy efficient since it generates GoP-size data bursts. Such behavior is quite different from that observed in previous studies for other applications, where EEE showed worse performance. Hence, video-streaming servers do not need any further technique to improve Energy Efficient Ethernet such as the aggregation of packets before transmission [6], since the video-streaming codecs already implement this functionality as part of its encoding process. It is also worth remarking that this study has focused on UDP-based video streaming, where it is reasonable to assume that the original packet timing is preserved for transmission. Future work will address the case of TCP-based streaming, where the frame transmission timing is expected to be throttled by the TCP congestion-control mechanisms, causing a significant power efficiency decrease. In addition to this, the TCP acks sent back to the video source are expected to further degrade the efficiency of EEE direction.