Prepare For Power Struggles In The Data Centre

Peter Judge has been involved with tech B2B publishing in the UK for many years, working at Ziff-Davis, ZDNet, IDG and Reed. His main interests are networking security, mobility and cloud

Low energy servers, and high voltage outages. It all add up to an industry power play, says Peter Judge

Two stories this week underline the importance of taming electricity in data centres.

Yesterday, we had the launch of HP’s Project Moonshot servers. The interesting thing here is a new architecture for data centres, based on microservers – and the big selling point is lower energy use.

And late last week came news that Amazon Web Services in the Eastern US region suffered a power outage.

Power outages

Coverage of Moonshot has focused on possible arguments behind the scenes.  Intel and Arm (with its Calxeda partner) are positioning to be the meat in the Moonshot menu, and much focus has one on the apparent slight to ARM, of HP picking Intel’s Atom for the launch.

It is possible that ARM servers are a bit radical for HP, but the Silicon Valley giant’s decision is more likely down the simple fact that Intel has more experience in data centre infrastructure, and it is taking a little longer to repurpose ARM designs (which rule in smartphones) into a full-blooded server (or even a microserver in a cartridge).

In any case, the main point long term, is the fact that, after years of server makers talking about lower power technologies, we have a new server architecture designed (or at least marketed) on the strength of lower power usage.

This is because power supply has become a limiting factor to the density of server rooms and to their location. Getting electricity to the racks has become a crucial issue and leaner designs could win over users.

The fact that electricity supplies are crucial came across all the stronger, as a lot of social media sites hosted on Amazon Web Services disappeared for a while late last week. Amazon’s Northern Virginia data centre lost power.

The cause here was a multiple failure. First, the power utility’s high voltage distribution system had a cable fault. Amazon was prepared and switched over to generator-based backup.

Unfortunately, there was a defective cooling fan in the primary backup service and one of the generators overheated. The system failed over a second time to a completely separate secondary back up system.

But bad luck struck again and Amazon had set the circuit breakers wrong in the secondary circuit. Their threshold was too low, so they opened at a normal voltage when the load transferred, leaving the centre’s servers with no power.

Once again, there is a short term media story: this time it is about blame. Amazon should have tested its primary and secondary backup circuits, and it has acknowledged that.

But the big issue underneath this is that energy is provided by a fragile infrastructure and measures to provide it reliably are expensive.

The more you can cut your power demands, the less costly the bills are and the less effort you have to put into providing reliable backup. Get the energy needs down low enough and you could think of living the dream of going off-grid, as HP has suggested elsewhere with its Net Zero data centre concept.

There’s no direct link between Moonshot’s launch and Amazon’s crash, but we are all on a trajectory which might avert future outages by simply wanting less power.

How’s your Green IT knowledge? Try our quiz!