I was recently a technical reviewer of a paper by the Greening of Streaming folks, called “Overview of Power Factor in Streaming“, and the there’s a few takeaways that surprised me enough to want to make a note to come back to it in future (here’s the linkedIn post announcing the paper). Power factor was something I was only dimly aware of, and I’m glad I have a better understanding of it now, as it has a few implications for discussions about the energy usage requirements of the internet. It also serves as a good example of how simple mental models we form about the steps we can take to reduce environmental impact can lead us astray.
What does the paper cover in the first place?
The paper introduces the concept of power factor when trying to understand the energy requirements of media systems like internet video streaming setups, how it can distort figures. It then provides some useful pointers for policy making.
First, introducing power factor
To explain power factor, I’m going to refer to a post I found helpful understanding it myself, from the Engineering Mindset website explaining the concept, with some hopefully memorable images.
They use a glass of beer as an analogy to explain what true power is, what reactive power is, and finally the ratio between them, the power factor.
Here’s the key quote:
The beer represents our true power or our kW, kilowatts. This is the useful stuff we want and need, this is what does the work.
The foam represents our reactive power or our kVAr, kilovolt-amps reactive. This is the useless stuff, there will always be some and we have to pay for it but we can’t use it so we don’t want too much of it.
Source: Power Factor Crash Course by Paul Evans
Power factor refers to the ratio between the two – so if a glass of beer was 50% foam, and 50% beer, you’d have a power factor of 0.5, meaning 50% is foam you can’t really make much use of.
Put another way, when you have a low power factor, the amount of supplied power you need is 1 divided by the power factor. So with a power factor of 0.5, the grid needs to generate twice as much power as you end up using, which means twice the fuel burned, and so on.
So, really you want your power factor to be as close to 1 as possible.
It’s a bit like how with servers in datacentres, you want a PUE (power usage effectiveness) figure to be as close to 1 as well, because the closer the you are to 1, the closer you are to all the electrical power going towards useful computation.
PUE isn’t the subject of this post, and I won’t go into how it is different – you just need to know that you the closer towards 1 you are the better.
Next, understanding why power factor matters with internet streaming setups
Where does power factor fit in to how we talk about the internet?
Well, power factor matters because not every device in a network has a power factor of 1, and depending on what that device is doing, the power factor can change.
Ok, can it change much? Let’s look at the paper:
- TVs: 0.9 when displaying content to 0.1 in standby
- Set-top boxes: 0.4-0.7 in active use, dropping to 0.1-0.2 in standby
- Network equipment: 0.5-0.8 load dependent
So, the short answer is yes, in relative terms it can – on a TV, the power factor when it’s displaying things can be nearly 10 times better than when it’s in standby!
This isn’t as bad as it sounds though, because a TV is likely using much more power anyway when it’s on. So, let’s assume you had a TV that used 40W when on, and because it’s an old TV, maybe 1W when it’s on standby. The power that needs to be generated by the grid is actually closer to a figure like 44W when it’s on, and more like 10W when it’s off.
With things like routers, the power factor varies based on how hard it’s working to send data over the network. Let’s take a quote from page 3 of the paper:
Base Load Dominance
Romain Jacob’s research revealed that many network routers typically consume 80% of their peak power consumption even during minimal activity. This high base load challenges the common misconception that power consumption scales linearly with data transfer. Even with aggressive power-saving features, consumption could only be reduced by 20-25% during low-traffic periods.
Power Factor Variation
Router measurements demonstrated how power factor varies significantly with load – excellent (0.95+) at full capacity but dropping substantially (0.6 or below) under light loads. This variation means equipment might require nearly twice the power generation during low-traffic periods.
Got that?
We might be assuming there’s a cars and driving relationship to sending data over the wire, where sending no data is like no driving and no energy being used. In reality, the power drawn at the device barely changes.
In fact, if the router’s power factor drops like a stone when it’s idle, then the power that needs to be supplied to it from the grid actually increases.
However, because power factor isn’t really visible to us, we’re totally blind to it.
What does this mean if you’re a well-meaning web developer, going to great pains to reduce the data sent over the wire, and thinking you’re halving the carbon footprint of a webpage by halving the data sent over the wire? It can mean that the mental model in your head is pretty much the opposite of what’s really happening on the grid.
It also calls into question what kind of interventions you can plausibly make when focussing mainly on code changes to an existing application.
Things like routers in homes are made to hit a specific price point, and one reason they have such wild power factor figures is that the special circuitry to make power factor better can increase the price of kit. The cost of energy used to power a router compared to the cost of the router itself also suggests that you probably wouldn’t replace the router. This is because for any improvements you see, you’d need to wait years before you see them become meaningful savings relative to the cost of the router, and that’s before you consider the energy and emissions that went into making a new replacement router too.
That said, in a datacentre or network exchange, the economics of this might be different. You might have a stronger incentive to make a change sooner, because it might “pay for itself” sooner, if the cost of energy over the lifetime of the service is greater, relative to the cost of the kit.
But broadly speaking, that’s not a decision you can influence by designing a web page all that differently, or deciding to skip a binge session on Netflix.
Why does this matter?
When designing models to help people reason about the actions they take to make software more sustainable, I think it matters in that it can help us understand why choosing a mental model is so important.
Simple models can distract people from meaningful actions, and when making policy, it can incentivise measures that are ineffectual and even counter-productive.
So it’s good that there are a few recommendations in the paper, and it gives some concrete advice when referring to policy. Here’s an example for Europe (useful in the context of the Ecodesign for Sustainable Products Regulation – the ESPR):
Current regulations often take an oversimplified approach. While well-intentioned, the EU’s 0.5W maximum standby power requirement doesn’t account for power factor effects. Due to poor power factor in standby mode, devices meeting this requirement might still require five to ten times more power generation.
If you’re coming up with any metrics or standards to guide behaviour around digital though, there’s a few in the paper:
- Develop standardised measurement approaches that account for power factor effects
- Implement repeatable measurement methodologies
- Consider power factor characteristics in equipment selection
- Focus on system-wide efficiency rather than component-level metrics
The paper is short and worth a read, and the Greening of Streaming folks publish a lot of good output on their publications page. I’m grateful for the invites to the workshops, as I learned a bunch from being involved.