< Back to previous page
Energieverlagende technieken in toegangsnetwerken van de volgende generatie
Book - Dissertation
Subtitle:Techniques to reduce energy consumption in next-generation access networks
THESE are interesting times for all of us. In recent years, our ecological awareness has dramatically increased due to better understanding, education, awarenessraising and perceivable changes in both nature and climate. Even in our daily lives major transformations are hard to ignore: energy saving lamps, electric cars, windmills, solar panels –to name just a few–, along with increasing energy bills. The Internet, and Information and Communication Technology (ICT) by extension, which has managed to penetrate nearly every aspect of our modern society in mere decades, is undeniably essential in the quest for means to reduce our ecological footprint and striving towards a sustainable future because of its unique capabilities to instantly and globally unite massive amounts of data, calculation power and specialized manpower. Thus, allowing to cope with increasingly challenging scientific and technological issues and to use scarce resources sparingly, whether by optimizing existing production processes, trading physical encounters for virtual alternatives or migrating from an economy based on the manufacturing of goods to one based on trading in knowledge and services. Though ICT clearly has the potential to evolve towards a sustainable society, a darker side becomes apparent. After decades of explosive expansion, the global ICT infrastructure has grown to consume a considerable share of the worldwide available energy. Though estimations vary considerably and are mostly based on –more or less conservative– extrapolations, alternatingly including office and consumer electronics, care is urgently required to avoid ICT from completely overshadowing its enabling effect. Among others, access networks and data centers are notorious for their power consumption. The introduction chapter of this dissertation reflects on the necessity of uninterrupted efforts and research to further increase the performance of the ICT infrastructure; on issues related to the rising energy consumption and on the considerable share access networks usurps. Besides, an overview of early attempts and initiatives to turn the tide is presented. The remainder of this work consists of two main components. The first component, described in chapter 2 aims to augment the functionality and performance of the present network technology. This work was a contribution to the IWT Vampire project, which suggested an access network topology in which a considerable part of the video processing which now takes place on the subscribers’ equipment –e.g. for digital TV and video conferencing– would be shifted to and centralized on specialized high-end hardware on the edge of the access network. This so-called “thinbox” concept, requiring mere basic and static hardware at the subscribers’ premises, at first would allow for more flexibility and smoother upgrades. In the opposing –“thickbox”–approach, new services all too often require upgrades of the end-user hardware, a statistical fraction of which ends in costly failures. The thin-box approach, additionally, could allow for a whole range of new, tailored services benefiting from the location in the network and reducing the data traffic to the end-user. The main objective of IWT Vampire was to define a novel hardware architecture for the delivery of enhanced video based services, and a number of technological building blocks (both software and hardware), necessary to enable these services in large scale deployments. The compressed nature of video streams, however, is an important obstruction to perform elaborate video processing. Consequently, research in the IWT Vampire project has focused on simultaneous and real-time H.264 encoding of multiple (10-1000) video streams. More specifically, Field Programmable Gate Array (FPGA) based building blocks for a high-throughput H.264 encoding architecture were researched. Focusing on intra-prediction, residual transformation and reconstruction, a throughput of over 13 million macroblocks (blocks of 256 pixels) per second can be achieved. This easily allows to simultaneously process 32 HD video streams at 30 fps. This was achieved by realizing a massively parallel and deeply pipelined architecture in which bubbles were mostly avoided by exploiting the availability of other video streams. This architecture is elaborately discussed in chapter 2. The second component of this work deals with a novel network communication protocol that could drastically reduce the energy consumption in the access network. This protocol, coined Bit-Interleaving PON (BiPON), is motivated and explained in detail in chapter 3. Its Application Specific Integrated Circuit (ASIC) implementation, coined BiCDR, is elaborately discussed in chapter 4. This new protocol specifically aims to reduce the energy consumption in the subscribers’ equipment. Because of their numbers, this equipment is responsible for a considerable fraction of global ICT power consumption. Passive Optical Network (PON), the physical layer for the new BiPON and the well-known B/(X)G/(G)E-PON, is a promising, passive, low-power optical access network technology which has seen massive deployment over the past few years.Its tree-shaped, shared nature requires some kind of multiplexing for the Optical Line Termination (OLT) to address Optical Network Units (ONUs) individually, despite broadcasting its message. To this day, only Time Division Multiplexing (TDM)-based approaches have been standardized and deployed, namely B/(X)G/(G)E-PON, which are packet-switched to cope with the typically bursty IP-traffic in the access network. In short, the problem with the current packet-switched TDM protocols is that each subscriber performs considerable amounts of processing on the downstream broadcast data at aggregated line rate, before selecting the useful data targeted for that specific ONU. Assuming 10 Gbit=s downstream, to be shared over 128 users which could roughly be assigned 100 Mbit=s each on average, data is processed at 10 Gbit=s while keeping a mere 1%. Clearly, this is a costly approach in terms of energy and reveals considerable room for improvement in terms of energy efficiency. Solutions relying on sleep-mode techniques have been suggested, yet Quality of Service (QoS) requirements and non-negligible sleep/wake transition times limit their efficacy. BiPON takes a radically different approach. A single frame contains data for all users, be it in a highly regular structure, while still offering high flexibility to scale and distribute bandwidth resources. The ONU can immediately subsample the incoming data, depending on the actual, dynamically adjustable, ONU-bitrate, virtually eliminating processing at the aggregated line rate. In chapter 3, a comparison between FPGAimplementations of both BiPON and XG-PON reveal an energy saving potential of a factor between 35 and 180, depending on the actual bandwidth allocated to a particular ONU. Chapter 4 presents the ASIC implementation of this protocol along with a Clock and Data Recovery (CDR) system and a decimator, which overcomes the well-known power inefficiency of the aforementioned FPGA-implementations. Chapter 5 finally presents the ongoing project DISCUS, part of the EU’s Seventh Framework Programme (FP7), which suggests radical changes in the entire network relying on state of the art optical network technology. Part of this project aims to take the Bit-Interleaving CDR/decimator (BiCDR) to the next level, namely 40 Gbit=s. Initial work is presented along with prospects for the future.
Number of pages: 1