Byte me

The energy costs of IT

I’m perpetually wondering about the environmental impact of my choices, and no, it’s not always easy being in my head. Lately, I’ve been contemplating what fraction of the energy used to stream videos is due to data transmission. As I was around a few computery folk last week, I asked them what they thought the energy cost of video streaming was. Here’s a bastardization of what they said:

1. Transmission of data uses a negligible amount of energy.

2. Data centers are the real energy hogs.

3. Bitcoin uses a huge amount of energy.

4. The internet saves us energy because we can communicate online.

After some google diving, I found a few more claims:

5. Watching 30 minutes of Netflix emits as much carbon as driving 4 miles.

6. The internet’s / Bitcoin's / data center’s energy consumption will rise exponentially leading to power outages and shortages in the coming decades.

How many of these statements do you think are true? After much stumbling around in the dark and reading conflicting news articles, I finally stumbled upon the International Energy Agency’s (IEA) reports on energy consumption. At last, real reports by energy experts, backed up by studies and presented with detailed units and explanations.

The first IEA report I read, soundly refuted the “30 minutes of Netflix viewing is equivalent to driving 4 miles” point. In turns out that the Netflix’s driving equivalency was reduced by the original authors by a factor of 8, when it was pointed out that they had interpreted Mb of data as megabits, and not megabytes. As there are 8 bits in a byte they’d overestimated the data stream by a factor of 8. Byte me indeed. Other errors also appeared into the original estimate such that 30 minutes of Netflix is equivalent to driving just 0.044 miles. This IAE article also broke down the energy usage of streaming a video into: 72% of energy used by the streaming device, 23% used in data transmission, and just 5% used by data centers. So this means the first and the fifth of the above statements are false.

So how about those data centers? In the last 10 years, although data demands on data centers have increased by a factor of 12 (or 1200%), their energy usage has only increased by 6%. This is because the energy efficiency of data centers has improved dramatically. The IEA predicts that continued improvements in efficiency will keep pace with increased demands for data traffic in the coming decade and gives examples of where practice can be improved and where new technology is expected. Data centers are hugely motivated to reduce their energy use as it represents 20-40% of their operating costs. So it seems that statement number 6 is wrong when applied to data centers.

In an eco-aside, data centers tend to be run by innovators who recognize that investment in green energy pays off in just a few years. What’s more, the localized nature of data centers lends itself to green energy solutions. This has led to data centers being the world’s largest investors in renewable energy. Both Apple and Google data centers purchase (or generate) 100% of their electricity needs from renewable energy sources. Facebook data centers use 75% renewables, while Amazon and Microsoft come in at around 50%. But like those of us who purchase energy from a green provider, purchased green electricity is not necessarily green, prompting Google to set a target of being truly 24/7 carbon free. This makes perfect sense when laid out before my eyes by the IEA, but it came as a delightful surprise to me.

The same IEA report that lays out data center energy use also documents and predicts rates of energy usage and efficiency for data transmission. Data transmission efficiency has also been keeping pace with increased digital traffic in the past decade and has an energy usage of the same order of data centers’, at just under 1% of global electricity use. However, being a distributed system transmission it is more difficult to run by renewables. So statement number 6 above is likely wrong for ‘the internet’ as well for it basically consists of data centers, networks (or data transmissions) and home and business computers. It may yet prove true for bitcoins, but I’m getting ahead of myself.

Before I delve into cryptocurrency, I wanted to address comment 4 above which I have hear repeatedly this past week - which is that the consumption of energy by the internet saves us energy because we don’t drive around as much. In the US at any rate, this is not so. Since the dawn of the the internet, the number of miles traveled has increase by about 3.5% every year from 1970 to 2020. Thus in the last 50 years the number of miles traveled has nearly tripled. There’s no indication that the internet has decreased our driving. Statement 4 false. Starting to sense a trend?

But of all the hyperbolic news stories, none is as extreme as those surrounding cryptocurrency. In 2017, newspapers reported on predictions that the electricity used by Bitcoin would surpass that of the US by 2019, and the entire world by 2020, and this from Newsweek no less. A Nature Climate Change article warned that Bitcoin alone could push global warming above 2 degrees C. Cryptocurrencies do use a great deal of energy because they rely on a book keeping ledger that is distributed across a network of computers. Each computer is part of a blockchain of software in which information in the network is passed on when a bitcoin miner solves a math puzzle. Some of the solvers are rewarded in bitcoins. The distributed nature of the transaction verification scheme makes it far more difficult to hack than standard centralized banking systems. I can only shake my head in wonder, but I can see that this distributed validation scheme means huge amounts of computing power for a given transaction because many Bitcoin miners compete to solve any given, computationally expensive, problem. It is apparently even more difficult to estimate total energy used by Bitcoin than to understand cryptocurrency, let alone predict its future use, but the IEA estimates that around 0.1 - 0.3% of global electricity is used for Bitcoin alone. There are of course other cryptocurrencies, as well as other uses of blockchain technology, so perhaps the whole cryptocurrency-blockchain conglomerate uses upwards of 1% of global electricity? The IEA states that we are unlikely to be energetically depleted by bitcoins nor is bitcoin likely to lead to 2 degrees of warming on its own. However, the IEA cautions that cryptocurrency’s environmental impact is potentially vast and worth keeping an eye on, thus statement 6 above may turn out to be true. Let’s hope not.

So we’ve seen that statements 1, 4 and 5 are patently wrong, and 6 is wrong for the internet but possibly true for cryptocurrency. So what of the claims that data centers are huge energy hogs or that Bitcoin uses an inordinate amount of energy? Data centers use about 1% of global electricity, data transmission also consumes about 1%, and Bitcoin consumes about 0.1 to 0.3% of global electricity. Is one of these an energy hog? Of the three - I’d say data storage and transmission have many values whereas Bitcoin not so much… But how about a sideways answer instead: home devices are the biggest energy hog. Scientific American estimates that around 5% of energy in the US is used to power computers so that’s (very naughty conversion ahead) about 10% of electricity. And if we simply stopped wasting energy on allowing our devices to idle, we could power Bitcoin twice over, or all the data centers in the world, or all the data transmission in the world.

Reducing energy spent on our devices does not require international agreements or technological innovations, rather, just a little effort on our part. A study in California found that on average we have 50 plugged in devices in our homes and 23% of household electricity bills is for “always-on” inactive devices. These idling devices in the US require the full time capacity of 50 large power plants worth of electricity running for the convince of not having to go switch something on when we need it. So to my mind, we’re the biggest energy wasting culprits. This makes our grand total of true statements at the start of this article to possibly 1 out of 6. Five of the statements are outright wrong. So much for the accuracy of fireside chats and shallow internet dives. But I’d best go. I think my printer is on idle and that would really byte.