Data Centres: Does Reliability Clash With Efficiency?

Computacenter’s new data centre in Romford got us interested at eWEEK. Partly because, well yes, we are that sort of people. But also because this is different from most of the data centre announcements we’ve seen lately.

Usually people call us up to tell us their data centres are green. They use less power, and they have a low PUE (Power Usage Effectiveness) so more of that power goes to the servers, and less to the cooling and lighting equipment around them.

Petroleum company GPS, for instance, has a PUE of 1.2 in its private Weybridge centre. Ark Continuity will have a PUE of 1.5 in its Wiltshire site, built in an old nuclear bunker and stone mine.

We’ve spoken at length with lots of people and worked on a Guide To Data Centre Management Efficiency – which we recommend you read, by the way.

And we’ve now heard of so many green data centres, we are beginning to think there is no other kind (or at least we were, until we heard from TelecityGroup’s Rob Coupland, who told us that so few companies have signed up to the EU’s Data Centre Code of Conduct that he is fearful the EU may come back with legislation). 

So Computacenter – and Infinity, which constructed the data centre – surprised us by talking about a different set of figures. The Romford data centre has been certifid by the Uptime Institute as tier IV – and is apparently the first of its kind in Europe.

Now, Tier IV is a measure of reliability, not efficiency (the details are in this PDF). What Computacenter is aiming at is a data centre which can provide a highly reliable service for companies that are so reliant on their data – they are willing to pay what Computacenter tells me is roughly a 20 percent premium.

Does reliability conflict with efficiency?

Computacenter and Infinity tell us the data centre is also efficient, but we asked them, is there a conflict here between the two goals?

Put simply, to make a data centre more reliable usually involves running more redundant hardware – including servers on hot standby. 

And in slightly more depth, we recently heard a Sun executive arguing that in some cases it’s possible to get more power efficiency, by running servers hotter – even if this reduces their working life. Hotter servers would be less reliable, so we’re pretty sure Computacenter won’t be following this rule.

After talking to the companies, we’re now pretty clear that there is at least some sort of trade-off between the two goals. But as it turns out, it doesn’t come out in these figures.

The Uptime Institute certifies facilities, not data centres, which include all the equipment in them. The Tier system operates at the level of power supplies and cooling, not at the level of servers. It doesn’t touch efficiency – except to the extent that more cooling and more back-up power may require more energy and more resouces (but not necessarily).

Likewise, the PUE figure misses out reliability – and even seems to miss out the efficiency impact of reliability. PUE measures the power delivered to servers, compared with the total power used in the facility. So if a server is on hot standby, the power it burns counts as power “used” not “wasted”.

In fact, an inefficiently used server, as in hot standby, might make a lower contribution to the PUE score, as it will generate less heat through things like disk reading.

Reliability and security is clearly another factor that data centres are going to have to address as applications move out of user companies into the cloud.

It may be that user companies have to meet requirements in both directions, and these may be backed up with regulations, particularly for companies in some sectors like financial services. Cloud compliance is another issue that interests us – enough to have run a web seminar on the subject.

In future, perhaps the two factors could be combined into an overall data centre fitness rating.