At the moment, regulators across the world are struggling to find ways to measure how energy efficient a data center is by measuring how the ‘useful work’ it delivers compares with the energy consumed. The idea being that data centers can then be given a ‘score’ which says how energy efficient they are – or are not – and the regulators can then influence the market.
Whilst I agree that data centers need to become more energy efficient, I think that in the quest to find one, all-encompassing energy efficiency metric, we’re being side tracked from the main issue and ignoring the big opportunity. For me, the elephant in the room is that data center environments have traditionally been designed, developed and managed as if their workload is static – and that is what we should be concentrating on first before we start imposing a metric.
To explain this, let’s draw a comparison between how a desktop PC or laptop uses power and how a data center uses energy. In recent years, chip vendors have worked hard to make laptops and PCs smarter in the way they use power – laptops and tablets in particular need a long battery life otherwise they won’t sell. This means that if you’re doing something quite complex on your PC – like editing a video or watching a movie – the PC draws more power than if you’re just browsing the web or working on a word document – and in standby mode, a PC hardly uses any power at all. Unfortunately, data centers don’t work like this.
Let’s say we have a corporate data center which hosts a customer contact platform with a web presence and a few thousand call center operators. On a Monday morning, it might be handling 50,000 calls and 100,000 web users every hour. It might be using a lot of power – but it’s doing a lot of work. However, if we return and measure energy consumption at 4am on a Sunday when there are no phone customers and only a few web users – it will be drawing almost the same amount of power.
To use an analogy, if a data center was a car we would not be able to turn the engine off and it would use almost the same amount of fuel idling on your driveway as if it were carrying five people, speeding along the motorway at 70 MPH. To continue the analogy, if you were concerned about your car’s fuel consumption – which you really should be – you’d probably ask why you couldn’t turn the engine off and why your car needed all that fuel when it was stationary. You wouldn’t ask the garage to tweak the engine to make it run more efficiently at 70 MPH – but this is the equivalent approach our regulators are taking in searching for a metric to measure how efficiently data centers run at maximum load.
For me, it is less important how efficiently a data center operates when working hard on a Monday morning. I am far more concerned that we understand how inefficient it is when it’s doing nothing – and that we then minimize this inefficiency. However, to date, the data center industry has a poor record of coming together to address this challenge.
Whilst some companies have recognized this issue and started making more energy efficient servers, much of the equipment in a data center still draws close to full power all of the time. There’s no one single piece of equipment that’s going to solve this issue and no one vendor or department that is solely responsible – there’s no magic bullet. To address this issue we need a paradigm shift in the way that we as an industry think about the design and deployment of data centers. From the mechanical engineering to writing our software, we need to ask: “how much power, data center capacity and cost will this waste to do nothing?” And as we ensure every device is tuned to manage its energy consumption at a lower load, we’re also going to see data centers becoming cheaper to run and more efficient at full load.
But to spur the industry into action, we need to create an informed market. We need to make the issue of waste energy and cost clear to the business. Every finance chief needs to know how much energy each data center service is using when it’s working at full capacity as opposed to when it’s idling for them to be able to make a decision which service is worth the cost – and which isn’t. Only then can they make an informed choice about which data center services they are happy to spend the money on, which ones they might want to spend less on and which can be turned off altogether. And it’s only when this happens that the market will start to change.
Interesting article raising concerns I hadn't previously considered. Machines running 24/7 use power and air conditioning meaning even more power even when underutilised. One way out if the impasse is to move more rapidly to shared resources - i.e. the cloud.
ReplyDeleteGreat article.
ReplyDeleteIn my opinion comprehensive 24/7 monitoring of the data center resources, form the physical infrastructure level (let's say: UPS utilization, managed PDUs monitoring - like how much current is provided to the given PSU etc.) through network bandwidth utilization, CPU power and memory usage up to the top IT infrastructure layers (databases, applications) and even higher is required. It would be great to know how many - for example - newly created insurance policies were handled during night batch processing and how it was correlated to the power consumption registered via all integrated monitoring systems. And so on... It can be related to any other business - not exactly insurance... One can imagine any other example.
Then - we all can agree - data center is not a static ecosystem. There are periods of higher and lower loads - with appropriate monitoring we can gain this knowledge. This is the first step. In the second step, we can learn how to optimize the resources over the time, including the requirements for both on-line and off-line (batch) processing, both on daily or monthly basis and so on.
I don't want to say that it is simple. It's not. However, I believe this is possible and the goal can be reached by integration of separate monitoring systems which are working on all layers of IT infrastructure - from physical infrastructure up to the applications level. Once we learn how our data center operates, we will be able to tune it for the real IT demands. Therefore, our business will be satisfied, as all demands are related from the higher to the lower level.
The only thing we have to change today is to include data center more into the IT departments and less into the facilities. Integration and convergence is the future of flexible and wise resources utilization. We shouldn't avoid the subjects like energy efficiency, power or cooling in the IT domain - our area of responsibility doesn't end somewhere in the layer of network/telecommunication systems. All IT infrastructure layers are based on physical infrastructure (power, cooling, space), therefore we cannot skip this domain from our scope. Otherwise, we can forget about efficient resources usage.