The starting basis for ECONET will be a heterogeneous set of energy-aware hardware technologies available
on the market and generally employed to build current network equipment (e.g. FPGA, ASICs, network/packet
processors, multi-core COTS processors and NICs). The project will explore how to effectively exploit and
adapt their power management features and capabilities in device architectures in order to meet network
operational constraints.
Fig 1: Estimate of power consumption sources in a generic platform of IP router.
To this end, this research axis will focus on the data-plane of network equipment, since this usually
include the most energy starving HW elements, with a weight equal to 54% vs. 11% for the control plane and
35% for power and heat management (Fig. 1).
Starting from this basis, the ECONET project will introduce, explore and develop two main kind of
network-specific energy-saving capabilities, i.e. dynamic power scaling and smart standby, respectively
(Fig. 2).
The first type of mechanism will allow network devices tuning dynamically the trade-off between energy
profile and processing capacity of internal processing blocks/engines, while meeting the actual traffic
load and QoS constraints. This will directly impact:
The design of novel traffic handling policies and queuing/shaping disciplines, which will be able
to effectively exploit active/idle HW transitions;
The development of green extensions for L2 protocols.
In detail, focusing on transport/core devices, modular systems are usually built with embedded switching
capability that is tuned for maximal system capacity. When the system is under partial load, then parts
of the system switching logic can be powered off or throttled in order to reduce overall system power
consumption without an impact on the overall performance.
For example, MLX modular switches system architecture relies on line cards and spine cards. The spine
cards provide the switching capability between all the line cards. The spine cards power can be adjusted
when the system is under partial load by either throttling speed/power of the spine cards or by powering
on/off complete spine cards according to the load. With these mechanisms the overall system power can be
adjusted. This can be controlled using interfaces, defined by the ECONET consortium, to overall
management of power of each system and of the overall network. Further performance/power trade-offs can
be also made available through the abstraction layer.
Fig 2: Green enhancements and capabilities at the device data-plane.
As far as the link protocols are concerned, among the other efforts in this field, the ECONET consortium
will study green extensions for lossless Ethernet (i.e., 802.1Qau, 802.1Qaz and 802.1Qbb). The properties
of such protocols can contribute to power reduction/saving across the network by maintaining smaller
buffers and avoiding energy investment on retransmissions. Part of the ECONET project can include research
of applying lossless techniques to the power saving schemes in the ECONET network equipment.
The second typology of mechanisms will allow putting currently unused parts of a network devices (such as
redundant network interfaces, unused network terminations, etc.) into very low energy consumption modes,
where only some basic functionalities are performed (e.g. heart-beating message reply).
The development of such network-specific low-energy modes is a fundamental key factor for disruptively
reducing the carbon-footprint in different network scenarios, since it will allow switching some portions
of the network (links, part of network devices, entire network devices) to a sleep mode in a smart and
effective way.
The key advantage of such solution with respect to a simple switching-off consists in reduced recovery
times as well as in the possibility of avoiding useless signalling storms of routing protocols at each
active/sleep transition of links or nodes.
In this respect, on the network access side, today most operators do not use low power mode available
on ADSL systems as entering and leaving this mode risks de-synchronizating other lines in the bundle.
The resulting re-training of the lines needs tens of seconds, which is not accepted by users.
Although optical technologies will not have these problems, millions of DSL lines will stay in place or
many years. Also with the mixed opto-electrical technologies like FTTC the last 100 meters remain copper
lines operated with VDSL. This technology offers more bandwidth at the expense of more power. It is even
more important to save power with VDSL.
Also in VDSL the crosstalk is higher, especially for the high frequencies. A special effort of the ECONET
project will be devoted to these aspects, and contributions will be focused on DSM mechanisms for VDSL
systems (DSLAMs and home gateways). For example, LQDE will bring in one or more prototype systems with at
least 16 DSL lines, which support echo cancellation in up- and downstream direction.
The idea exists that in a fully equilibrated DSM bundle there is virtually no crosstalk. Hence a line of
the bundle could be switched off and on again without disturbing the other lines.
