It's Time to Make Streaming Greener
In the online video industry we spend much of our time facilitating the entertainment of the general public. We encourage "engagement time" on our and our clients' services by making better content and delivering it with excellent quality when and where the users want to consume it.
We're also consumers. Whether it's binging on our latest video obsession, snacking on fast food, driving cars, or relying on battery-powered devices—not to mention heating and lighting our homes—we do dozens of things every day without giving them a second thought. We hope that behind all of those products and services are teams focussed on ensuring that they are manufactured and delivered in ways that are sympathetic—or at least not overly destructive—to our environment. We recycle. We choose carefully. But ultimately we assume someone else has taken responsibility for making sure we are not unnecessarily killing our planet as we do so.
But as the creators, facilitators, and enablers of all this video consumption, we are the ones who should be shouldering the responsibility to make sure that our supply chains are as environmentally sound and sustainable as possible. This is not someone else's problem. This is our problem, and the solutions should be ours as well. Whether you want to save the planet, or you want to save money and return more to your shareholders, the interests are aligned.
Streaming video and audio use about two-thirds of all network services today. Those network services are complex beasts—where our decisions can make huge impacts.
To paint a painfully simple picture, by some estimates today the internet as a whole is consuming somewhere between 7% and 10% of all electricity, and that is likely to double in the next 5 years. Estimates vary depending on whether the figures are published by power and fuel companies, external agencies, the UN, and so on, but they don't vary too widely. Also note that around 82% of all energy is used on fueling heating and transportation, which indicates that the internet is consuming about 2% of all energy.
Let's look at that in tangible terms: Around 1 in 50 coal trucks, oil barrels, rotations of wind turbines, square meters of solar cells, or one part in 50 of all water flowing through a hydropower station is generating the power that creates the network that our content uses to reach consumers.
That proportion will grow to 1 in 25 by 2025, if estimates are correct.
World energy supply in 2017 was ~25,000 Terawatt hours (TWh), and if ~20% is used for electricity, and if 10% of that electricity demand is the internet, then the internet demands 500 TWh annually.
But how about streaming?
We have to start somewhere, since this is not a question those of us in our industry have typically asked of ourselves. And we will have to make some assumptions: My starting point is to assume that there is a direct correlation between the power demand by application and the traffic demand by application. Engineers will cringe—my apologies—but this is a worst-case assumption and therefore a useful boundary.
So using the Sandvine number that 62% of mobile internet traffic is video, we can work up a back-of-envelope estimate: video streaming is using about 25x0.2x0.1x0.62 = 310 TWh annually.
The U.S. has an annual power demand of around 4000 TWh.
Put another way, to deliver to the global video streaming demand today creates an energy demand equivalent to 8% of the entire U.S. energy supply (including heat and transportation).
With a global estimate of the renewable component of our energy supply sitting at around 6000 TWh this indicates that global streaming could be consuming as much as 5% of all renewable energy.
And to make one final comparison, U.S. nuclear power accounts for about 800 TWh from 98 power stations. While acknowledging the margins of error in the estimations I am making, I think it is fair to claim that global streaming video delivery today consumes in the magnitude of the power of 50 nuclear power stations!
OK, these numbers are designed to shake you up a bit. The point is not the accuracy—in fact, I welcome corrections—but the orders of magnitude.
Notice I have been careful not to talk about carbon. Carbon has become politicized. I don't seek to make a political argument here. I am lobbied by neither petrochemical companies nor by Greenpeace.
I am simply painfully aware that we, as an industry, are not discussing power efficiency at all. And inefficiency in any form is wasted value. Regardless of your personal beliefs around politics, power, and the energy markets, your shareholders will thank you if you can make your workflows more efficient at every level.
And it's important to note that if we in the industry don't take the lead, we're only going to see more criticism from the outside, as with the recent controversial BBC documentary Dirty Streaming, as well as less sensational investigations like a 2019 study on the environmental impact of streaming music services.
Lowering Ingest and Transcoding Power Consumption
While the improvements per bit have been slowing somewhat (like an inverse square law), codec compression has been improving year-on-year over the past decades, gradually reducing the need for increased network capacity (and of course the power it requires).
CPU processing speed increases have slowed now, but GPU is increasingly being leveraged, boosted by adoption in the crypto market. FPGA is also emerging fast as an option. Indisputably GPUs, FPGAs, and ASICs are more power efficient for encoding a frame of video than CPU, so strategies to adopt GPU and FPGA in encoders are beneficial, albeit at the cost of the flexibility that we have grown accustomed to with CPU.
Always on/instantaneous delivery comes with a power requirement. CDNs are always working to optimise caching strategies. But one has to keep a check on the fact that the caches are hot all the time. Yet if the content is neither cost- nor latency-sensitive, does it need caching at all? A single origin may mean a slower delivery to the end user, but if they don't mind, then why spin up thousands of edge caches and increase the energy demand 1000-fold to deliver something that does not require that caching at all? Do we as an industry ever ask these energy-related questions in our pursuit of QoE, or is that question simply conflated (and obfuscated) by the effect on service price alone? In 25 years working with streaming technologies, I have rarely if ever had this discussion with anyone other than data center and internet exchange managers.
In talking with colleagues at Streaming Media I discovered some strongly shared excitement about this discussion. Some time ago Contributing Editor Tim Siglin carried out paid research on the massive power consumption of DVRs, set-top boxes, and streaming devices and so he has long fostered an interest in the topic; and by the nature of his role as Streaming Media editor and VP, Eric Schumacher-Rasmussen has a broad interest in the industry as a whole. So going forward Tim, Eric, and I—along with (hopefully) an ever-growing community from within our sector—are going to start beating this drum louder and louder here and in the pages of Streaming Media magazine.
The numbers in this article are all cursory results from preliminary research. The aim now is to refine that data set, decide what data is really relevant, and work with the community to formalise a consistent set of measures with which streaming companies can benchmark, and work to improve on.
Naively, we have evolved nearly three decades into streaming without a public discussion and focus on these issues, and it's high time our sector really comes to grips with what we've done, what we're doing, and how we can change for the future. Indeed, as more people are working from home, using videoconferencing, and watching OTT video, there's never been a better time to get the discussion going.
So please get in touch if you can get involved.
This year we are going to open the discussion about establishing a "Streaming Media Green Index," leveraging our industry's deep technical insights to start asking some (probably difficult and painful) questions about streamlining our streaming architectures, not just for QoE and price as we always have, but for power demand, pollution, and misuse of non-renewables.
Like most of you, we are streaming industry experts but not necessarily experts (yet) in measuring the green values of all the elements in the workflows we explore. This will hopefully be a collective/crowdsourced effort so regardless if you have experience in this space, or perhaps can contribute pointers to those who advise you already please get in touch on firstname.lastname@example.org
Let's start this overdue discussion now.
The time has long since passed to turn our attention to making the streaming delivery ecosystem more environmentally friendly. Here are some of the key challenges, as well as some suggested first steps towards solutions.