[Tech Talk] You Can't Survive Without Compression

Internet Filled with Video Images
More CDN Investment, Inevitable Power Consumption
Unsustainable Without Stronger 'Compression'

Global concerns about the electricity consumption of data centers have been growing recently. To meet the demand for artificial intelligence (AI), the global power demand for data centers is expected to be three times higher than it is now.

If high-resolution images are not compressed, the cost of maintaining internet infrastructure could increase astronomically.
[Image source=Pixabay]

However, before AI computing, the real culprit behind the internet becoming an electricity-guzzling beast is something else. It is the images and videos we enjoy daily through our computers and smartphones. These two consume much more capacity and data bandwidth than text. Moreover, as the quality of videos and images improves, the electricity consumption for internet usage must increase exponentially.

'Express Delivery Service' Sending Photos and Videos to Our Homes: CDN

To quickly deliver large-capacity content to users located far from the data center (orange), a separate network (blue) must be installed in each region. This network is called a CDN. [Image source=Cloudflare capture]

The digital world is already filled with photos and videos. Just by entering a portal site, high-definition photos and videos quickly fill the screen. When accessing video streaming sites like YouTube and TikTok, you can watch vivid videos in real-time from anywhere in the world. How have we come to enjoy such conveniences of civilization?

The key is the Content Delivery Network (CDN). A CDN refers to nodes that quickly deliver large volumes of data (i.e., images or videos) to service users.

People often say that today’s internet is made up of clouds centered around data centers. However, data centers alone cannot deliver massive amounts of data directly to our home computers at once. Bandwidth shortages would cause bottlenecks. Instead, various CDNs are installed between the service provider’s data center and the user’s PC to smooth the data flow, which is how the current internet operates.

Without More Powerful Compression Technology, the Internet Could Face Overload

Thanks to the development of cameras and video recording devices, ultra-high-definition photos and videos have become commonplace today. [Image source=Yonhap News]

Thanks to this, we enjoy an internet based on videos and images, but the problem has not disappeared. CDNs have rapidly grown since around 2008, when the era of social networking services (SNS) fully opened, and now account for a significant portion of big tech companies’ annual business investments. For example, Meta (Facebook) is known to spend over 10 trillion won annually on CDNs alone.

However, the quality of photos and videos will continue to improve. This means the costs and electricity consumption required for CDNs will also increase. How can this vicious cycle be stopped?

Until now, the solution has been "compression." The most widely used image file extension standard today is "JPEG." JPEG is a technology that compresses image size using algorithms without significantly damaging image quality.

The widespread adoption of the JPEG extension has reduced the burden on internet bandwidth.

Most photos shared on the internet today are in JPEG format. It has become so common that people use JPEG unconsciously, but without this compression technology, the "imaging" of the internet would have taken much longer.

However, existing algorithmic compression technologies have clear limitations. Especially with the development of video streaming technology, the load on CDNs is reaching unprecedented levels. Ultimately, more powerful compression technology surpassing JPEG is needed.

In the Streaming Era, Could AI Compression Be the Savior?

Ironically, AI, another electricity-guzzling beast, is being hailed as the next-generation compression technology. This method involves converting images or videos into low-frequency bands and then entrusting the process of converting frequency-based files back into data to neural network intelligence. Unlike human-made existing algorithms, it can compress data to an extreme level.

DeepMind's 'MuZero' AI compresses and transmits videos.
[Image source=DeepMind]

The company that first perfected this technology was DeepMind, Google’s AI research subsidiary famous for AlphaGo. Using another neural network intelligence derived from AlphaGo called "MuZero," they extremely compressed YouTube video data, resulting in a 4% reduction in traffic for YouTube’s data centers and CDNs in 2022.

Although 4% may seem like a small number, even a slight improvement is a great help in the network world, which boasts enormous electricity consumption. The bandwidth saved by suppressing traffic can be invested in other services or used to provide improved services to users at the same cost.

Other big tech companies are also putting great effort into exploring AI compression technology. For example, Apple acquired a startup called "WaveOne" last year. This company is developing AI algorithms for video compression, which may be used in Apple’s streaming services such as Apple TV+ in the future.