How much less power does a data centre consume when it has lost its internet connection? No traffic in, no traffic out, no serving of websites, processing of queries, no email. Do they use less power? Regretfully I have recently been in the situation where I could find out. Twice. On both occasions, this was due to circumstances outside of our control*, and in absolutely no way anything to do with my usual blundering manner. So - let's go over this in a little more detail with specifics.

// What are we talking about here?

The concept or hypothesis is thus:

"How much more power does a racked enterprise server environment consume when delivering active load compared with when they are idling, and simply doing their homework. One would assume, noticeably less."

// Reasoning

Given that machines will ramp up their fans in an attempt to cool themselves when a load is sufficiently high to cause heat build up - this is going to have implications directly. It will also have an indirect impact as this heat will be flown through the (poorly conducting) medium of air into the facility - which will then need cooling to ensure that cool enough air is being drawn in to maintain temperature. You can expect the cooling cost of any facility to be around a third of the power consumption. That is to say that two thirds will be your computing load, and a third will be your cooling.

// Example

Back to this specific example - conveniently a monitored sub-main delivers 3 phase to a unit of 24 racks. They are not entirely populated. The load is decidedly mixed. Shared hosting. Shared email. Dedicated servers. Virtual servers. Very few are in the habit of ramping up meet their cooling to meet their needs. The majority of traffic is outbound. The majority of web traffic being 'small requests in, and websites back'. Email is far more give and take. Admin traffic, negligible. Machine housework is going on as usual.

// Findings

So - does it change?

Well I am confident there will be the odd Amp/Watt here and there - however predominantly when measuring in power units - NO.

There is no noticeable change in the consumption with or without a connected internet causing a mixed load of tasks, storage, processing.

Obviously 'Your Mileage May Vary' and all that - I guess it very much depends on what you are doing with your internet traffic. Something hardcore transactional or computational may differ - but for web and email traffic. Nada. Nothing. Barely noticeable. Not enough to confirm there is a noticeable difference from the norm.

So. Now we know that - let's not find ourselves in a situation to test again for a very very long time.

* Geo diverse, multiple feeds, pinch points, 'national provider skillz'.