Data centre power consumption is under the spotlight.
Energy efficiency is one reason for this, but the more pressing issues are capacity, continuity and cost.
Many centres are at their limits in terms of available space and also the amount of power they can draw.
To keep pace with computing growth and ensure continuity of IT services, most IT and centre managers are looking at ways to use power more efficiently and extend the life of their facilities.
Racks – and the data centres that house them – have a finite power envelope, based on in-rack and -room thermodynamics and the physical capacity limits of the grid.
Yet as user growth demands increase, we seem compelled to try to pack more into both.
Rack density has spiralled in recent years. Depending on who you speak to (and believe), seven to 15 kilowatt (kW) densities are bandied around as commonplace – and rack vendors such as Rittal believe seeing 40kW densities aren’t altogether out of the question.
Miniaturisation of the internal server electronics is partly to blame.
It makes smaller but more powerful architecture configurations such as blades possible.
It’s no secret that these types of servers also run hotter and require more cooling (and therefore use more power) than your average volume server.
The addition of more communications equipment into enterprise environments is also contributing to densification.
Network switches and other equipment used to enable Unified Communications, VoIP and video-over-IP are becoming mainstays in the data centre environment.
This appears to be placing further strain on already limited power resources.
The industry is largely in agreement on where power management in the data centre needs to go.
It’s all about measuring the power draw of equipment at the socket or node level, and using that as a baseline to reduce electricity consumption and costs.
Measurement, it could be argued, is the easy part. The challenge for IT managers is what to do with those metrics.
FM versus IT. Fight!
A bit of background first. A philosophical argument is currently raging among vendors over who owns the power bill internally in organisations and as to whether or not facilities managers (FM) and the IT folks have come together to collaboratively reduce power consumption, increase energy efficiency and cut costs.
After all, if CIOs and IT managers don’t see their electricity bill, is there a motive for them to take steps to reduce consumption and costs?
Vendors say yes – for corporate social responsibility if nothing else – but in the industry, opinions are mixed.
Peter Spiteri, senior marketing manager of Emerson Network Power Australia, is adamant that change is here.
“The way to shape behaviour is organisationally – by bringing the IT department under finance [instead of under the CTO],” said Spiteri.
“Large organisations including Emerson have done this. In the past three years we’ve moved all our IT reports up through the finance department.”
He claims that for some CIOs and IT managers, as much as 50 percent of their role is now mandated towards increasing energy efficiency, in addition to maintaining system uptime.
But other commentators are not so sure.
“We’ve just been speaking to Australia’s second or third largest telco and found neither party [FM or IT] was responsible for proactive monitoring of the data centre power consumption,” explained Gary Hull, director A/NZ at Raritan.
“It’s a scenario we see and hear over and over again. Finance is responsible for paying for power, facilities for provisioning it and for continuity, and IT for service delivery and uptime”.
Hull does, however, see the responsibility starting to fall to IT.
“Facilities are typically the UPS level of power. They look at how much power is delivered to the data centre and don’t get visibility beyond that,” said Hull.
“As the [above] scenario changes, that to us presents a real opportunity.”
“It’s easy to get an understanding of power usage at the facility level, but customers now want to understand it at the IT level,” claims Mark Roberts, business development manager for IT at Rittal.
Roberts divides the data centre power load in two – a perspective he calls the ‘two sides of the house’.
On one side, the facility load is powering airconditioning, chillers, uninterruptible power supply (UPS), power distribution units (PDUs) and so on, while on the other side the IT load draws power for servers, networking equipment and storage.
“You obviously need to monitor both sides,” explained Roberts.
“Traditionally most people have monitored power distribution high up at the facility level, but now they want to monitor closer to the actual piece of IT equipment, which means monitoring at the socket level.”
Socket-level (also referred to as node- or equipment-level) monitoring is this year’s hot button topic in power management.
“Taking measurement down to the socket level is a major step forward for organisations from a visibility standpoint,” said Raritan’s Gary Hull.
“Measuring by node or infrastructure device means customers now have a benchmark to gauge power usage and can work towards reducing consumption and costs from the baseline they’ve established.
“It helps them understand what devices are consuming power, how much are they consuming and is it necessary for them to consume that much,” explained Hull.
Hull said Raritan had seen a strong understanding demonstrated by local data centre operators of the need for node-level monitoring and measurement.
That comprehension had already translated into sales into the Department of Treasury and Finance and JP Morgan, among others, he said.
Ciaran Bolton, product manager for data centre solutions at Eaton, is also focused on socket-level opportunities.
“It’s really coming right down to the node level as IT managers want to get as granular as possible,” Bolton said.
“We’re getting to the point where the cost of energy is almost the largest single cost in building or running a data centre.
“If you can meter and monitor down to the node level, you can identify and consolidate volume servers running at 10 or 15 percent utilisation to stop wasting that energy and save some money,” Bolton said.
To understand why socket-level measurement is so important is to take lessons in the disparate worlds of rough calculation and thermodynamics.
According to Rittal’s Mark Roberts, IT managers typically work out how much power they’re using at a socket level by adding up the plate ratings listed on the back of each piece of equipment.
They then subtract this total from the room’s known capacity to work out the power window they have left.
“The plate rating is usually the worst case scenario – the true working scenario is a lot less,” explained Roberts.
“For example, if the rating says the box draws five amps it’s probably only drawing around 2.5 or 2.75 amps in reality.”
Roberts said he routinely applied what he calls a ‘diversity factor’ of 0.6 to listed plate ratings for this reason.
However, measurement can (and in the current environment, probably should) be a more exact science.
“If people have a way to dynamically measure power, they can see exactly how much each box is drawing,” said Roberts.
Gaining a better understanding of the socket-level power draw can result in exponential savings at the facility level. Emerson refers to this as the cascade effect.
Put simply, the cascade effect says that every one watt of savings made at the server component level (processor, memory, hard disk etc.) creates a reduction in facility energy consumption of approximately 2.84 watts. (i.e. a one-watt reduction in IT load plus a 1.84-watt reduction in the facility load).
“When you take a single kilowatt off the grid and into the data centre you get losses all the way through [to the IT equipment],” said Spiteri.
He added: “Around half the power that comes in is used to power and cool the building as opposed to power the IT equipment. Right upfront only half the power you buy is going into the whole reason that the building exists.”
This is the key to selling socket-layer measurement – it helps gain a better understanding firstly of how much power each box is drawing on and secondly how this affects the total power consumption of the centre.The main way to take a measure of power consumption is directly from the power distribution strip.
It is possible to measure at both a strip and socket level, however the underlying technology in the strip for each type of measurement is different, and the more granular you want your measurement, the more expensive it gets.
“When you measure at a power strip level you only need one or two current transformers to measure the power,” said Rittal’s Mark Roberts.
“If you want to measure at the socket level then you need a current transformer for each individual socket, so the technology required grows with the granularity.”
Ballpark figures for power strip-level measurement hover around $800-$900 per strip, Roberts said. Go down to the socket level and it jumps to $1400 or $1500.
A large data centre packing 500 racks would need approximately twice that number of strips, each of which might contain eight or 10 sockets.
“For that sized centre it could cost about $1 million to monitor at an individual socket level,” said Roberts.
Raritan’s Gary Hull agrees that intelligent systems are the answer, but at the PDU level.
“What Raritan has done is provided the market with an intelligent PDU that measures, meters and switches power by node or infrastructure device,” said Hull.
“This provides a great opportunity for partners to offer proactive power management solutions that augment our [traditional] infrastructure management systems.”
The company also has open source software called Power IQ that can consolidate active power information derived from multiple PDUs at the node level.
Hull said it works with any device that is both SNMP-enabled and can event forward management information base (MIB) files to Power IQ.
Other vendors have similar software offerings, including Eaton and APC.
Like Power IQ, both of these solutions also plug into other third-party management systems such as IBM’s Tivoli.
Eaton is also selling intelligent monitoring primarily at the PDU level.
“We have products that can be retrofitted to ‘dumb’ PDUs so that metering or monitoring capabilities can be added to existing sites quite easily,” said Eaton’s Ciaran Bolton.
Both Eaton and APC are also proponents of taking a modular approach to UPS infrastructure.
Adaptec, meanwhile, provides a way for IT managers to start saving at the equipment level. The company has created a firmware enhancement for its Series 2 and 5 RAID cards that enables IT managers to either spin down or slow down attached drives to a low-power state or complete hibernation to save energy.
“Traditionally there were specific appliances to do this,” said Neil Cameron, field application engineer at Adaptec.
“Our product is for the generic market from workstations upwards. People can simply turn on these new features.”But technology is only one part of the solution to power management.
Choosing a framework in which to implement it is a key challenge.
The number of options available is staggering to say the least – which is why many customers will likely look to system integrators and service providers to help them choose.
In simple terms, the options are power audits, standardised data centre efficiency measurements, all the way up to complete frameworks for implementation.
APC offers a five-step methodology that covers convening an IT and facility workshop to scope issues, a power audit, offering engineering and design input, creating a 3D simulation of the revamped facility and finally project management skills if needed.
“We actually take in thermal cameras to take images of the racks, do some computational fluid dynamics around the airflows in the racks, and look at rack configurations and layout with specific tools,” explained Makryllos.
Makryllos said the IT/FM workshop, site analysis and conceptual design would typically cost around $4000.
Eaton offers a similar though higher-end service. The ballpark cost is around $10,000 up to $20,000 for ‘a real in-depth audit of the facility’, explained Bolton.
“Our dedicated engineering services team will come out and take a look at the customer’s entire power chain and conduct a safety and efficiency audit,” said Bolton.
IT managers may also wish to use industry-standard data centre efficiency measures as part of such an audit. Examples include power usage effectiveness (PUE) and data centre infrastructure efficiency (DCIE).
Emerson has also created its own in-depth power management methodology called Energy Logic.
It’s a ‘vendor-neutral’ 10-step program of best practices that can be implemented individually or altogether, and lists anticipated kilowatt savings and return-on-investment upfront.
In conclusion, measuring the power draw of IT equipment is one of the key actions that can be taken to reduce power costs. The trend to socket-level measurement will not only provide accurate energy consumption data, but also a baseline to reduce electricity consumption and save money.