Demand-side energy management starts with an energy
monitoring capability that requires no extra hardware to be installed. For
instance, Sentilla Energy Manager tracks energy consumption at a data center to
within 2 percent of actual, as well as provides savings of up to 40 percent by
identifying where energy is being wasted.
A Web-based dashboard allows IT managers to view the real-time power consumption
of the data center and pinpoint waste.
"We are treating energy as an IT asset, reducing power
consumption at data centers while simultaneously increasing capacity and
performance, but without the cost and disruption of installing metering
hardware," says Sentilla Chief Executive Officer Bob Davis.
Previous demand-side energy management software required
data centers to install utility meter-monitoring hardware on any unmetered
devices. But the new Sentilla Energy Manager (Version 3.0) instead uses an
artificially intelligent (AI) inference engine to deduce the power consumption
of unmetered devices Sherlock Holmes-style.
First, the open-source and proprietary agents in the data
center are queried by Sentilla Energy Manager to catalog each piece of
equipment in the data center, its configuration and CPU utilization. The AI
then looks up the typical energy consumption specifications for that type of
equipment from a database of equipment profiles and estimates its energy
consumption.
"The basic parameters for most data centers is how many
CPUs each server has, how many disks, how many power supplies, plus how much
utilization there is, all of which is fed into the inference engine,"
Davis states.
The second thing the AI does is build a hierarchy of the
data center. The inference engine refines its estimates for the energy
consumption of the unmetered equipment by the same method used by Sherlock
Holmes—namely, deduction. For instance, if one branch of the hierarchical tree
is using 100 watts, and it has two leafs, one of which is metered at 40 watts
and the other estimated at 75 watts, then by deduction (here subtraction) the
AI can infer a 15 watt over-estimate on the unmetered equipment, since the
unmetered device's consumption must be 60 watts (100 minus 40).
"If all we know is how many CPUs, disks and power
supplies a server has, then we can estimate how much power it is consuming to
within about 10 percent," explains Davis. "But when we add in the
adjustments to our estimates made by the inference engine using the
hierarchical model, then we can get down to about 2 percent accuracy, which is
just as good as adding a utility meter to every device."
In 2009, the Sentilla Energy Manager received an American
Business Award for Best New Product of the Year, the TechWorld Award for Green
Product of the Year, and the Best Technology Supplier of the Year for the
British Computer Society's IT Industry Awards.
