The power consumption of the Internet is very big and most of this energy is needed to power datacennters. For example, the supercomputer of BSC MareNostrum (which I visited two weeks ago) has a power consumption of 1.2MW which costs more than 1 million €/year, and it is not the most powerfull supercomputer in the world.
Because of the huge amount of energy needed, several measures have to be taken in order to optimize the servers and the cooling system of the datacenters, which are the two systems that consume more energy.
Virtualization is an effective way to save a significant amount of energy. We can offer different services in one physical server with virtual machines (virtual servers) which improve their consolidation and efficiency. It means economical and energy savings.
When servers are on idle mode, putting them on sleep mode could have important savings in terms of energy. The ideal case consists on combining sleep mode (downclocking, adaptive link rate, per-interface sleep mode, low power idle) with job aggregation (exploit traffic fluctuations to aggregate jobs on a subset of servers and turn-off the idle nodes).
On the other hand, network/site security must be considered. Attacks could change in their main aims, exploiting weaknesses in power-saving and management mechanisms to disrupt services, or even attempting to increase the energy consumption of an entire farm, causing financial damages. However, we can fix attacks by setting a maximum power consumption threshold and operating the facility always below. Moreover, if an increment is detected, downcloking the device, forcing sleep mode or scaling CPU/Voltage can decrease the power consumption.
Energy oriented approach, as we saw in previous post, is an effective way to minimize power consumption, GHG emissions and costs. Datacenter placements near green energy sources reduce dispersion, cooling and energy costs. Moreover, following the weather paradigm minimize power consumption, GHG emissions and/or energy bill.