Cost-efficiency of cloud computing. Centralized vs. decentralized networks.

If privacy is costly it won't be protected. How to enforce it with a decentralized confidential computing platform.

Currently, I’m looking into Ethernity CLOUD, a project that works on a decentralized network, that lets data be processed in a confidential, privacy-protecting way - by default. In order to understand the project's viability, I recently realized that it’s crucial to understand the cost-effectiveness of decentralized cloud computing networks compared to centralized ones. You’ll discover that this economic question lays the foundation for a moral purpose. A purpose described in my last article: Let's protect our online privacy by decentralized cloud services.

A key, full of rust, structural problems by the dollar bill texture.

This piece follows up on the topic by observing that if privacy is costly or not profitable, it most likely won’t be protected. (I tried to visualize that in the picture above). So the underlying question which needs to be answered here is: Can we incentivize privacy with a decentralized cloud computing solution?

First, let us understand the costs. When we’re using

• a streaming service like Netflix or Prime,

• a collaboration tool like Slack, Trello, or Miroboard,

• a social media platform (that sells our data to pay the bills)

• or mobile apps in general

what is it that needs to be paid for?

The easiest way to get a grasp on that is to distinguish between the expenses upfront, e.g. the investment in the infrastructure (CapEx) and the operational expenses (OpEx). Since cloud computing providers are typically companies, they have a high percentage of expenses in OpEx. Reasons for that are massive cooling costs, fire protection, facility management expenses, and support for their users. Furthermore, to be compliant with policies, companies have expensive legal responsibilities.

TL;DR: These are costs that would be way lower or redundant in a truly decentralized network for confidential cloud computing. I’m gonna explain why. But first things first.

1. Centralized Cloud Computing

Why do we even have centralized cloud computing service providers?

There were reasons why it made sense! As a company, a university, a hospital, or another type of institution you needed to buy, install and maintain your whole IT infrastructure mainly locally until the early 2000s. If you haven’t done it by yourself you just asked a service agency to do that. But there was a shift away from that. Marked by Amazon Web Services (AWS) in 2006 and under the shadow of the 2008 crisis, clouds emerged in full strength. Why? For many institutions, it wasn’t possible anymore to afford a local infrastructure, to risk downtime of their computers, or to keep up with the security standards. I recommend reading this well-described history of cloud computing for a more detailed perspective. 

In addition to that centralized cloud providers could also get bulk discounts for energy. Costs were distributed more efficiently. So switching to such a provider who takes care of all of these costs seemed to be the best option! This way you could just focus to “do what you love” - as a random company slogan might put it. 

Centralized cloud computing is more costly than expected. 

But as it turns out cloud users and cloud providers couldn’t save as much money as desired. For users, the switch from on-premise infrastructure to a cloud provider had turned out to be more costly than expected, yet it was perceived as something that was worth doing. 

For Providers, the previously mentioned operational costs are sucking up most of the revenue. In Amazon's case from $56.8B gross profit, they have $54.3B operating costs and a net profit of around $2-3B. Energy is one of the biggest portions of this calculation. For example with a price of 16 cents per KW/h (that’s an average for US businesses) Amazon - with its 30 TW/h consumption in the year 2021 - would have electricity bills of $4.8B if all the servers were in the U.S. That represents almost 10% of their total OpEx. You can find similar data visualizations here for Google, Microsoft, and Apple. 

So we’ve now lifted the curtain and seen behind the expenses of web companies that we are using daily. Centralized managed cloud computing platforms have 80-90% of their expenses in operational costs. Given the context that privacy was our main issue with these services, this shows us a crucial economical problem and lets us hope for an even better, more cost-efficient solution. A decentralized one. And it shouldn’t be something that is only made for the sake of decentralization but for the sake of making a privacy-protective cloud computing platform viable. 

numbers based on Arcitura

2. Decentralized Cloud Computing

But how can a decentralized network be more cost-efficient than a centralized one?

That’s a good question especially when you look at the often-stated potential downsides of decentralized networks:

• It’s slower. (a) because a decentralized blockchain needs to achieve consensus or (b) information goes through several nodes before it arrives

• It’s harder to scale them (e.g. the whole network has to agree for software updates)

• It’s more expensive (a setup for 1 server is less efficient than a setup for 1000 nodes) 

• It’s worse accessibility (Who do I reach out to if sth. doesn’t work?)

• It’s not truly decentralized

These issues can be summarized as the “blockchain trilemma”. Depending on the way how a network decides to approach this trilemma of decentralization, scalability, and security, it is stuck with 1 of 3 aspects unfulfilled. Furthermore 24/7 availability and security - possible strengths of decentralization - are at risk in the early stages of these types of networks. Even if these “birth difficulties” disappear with the growth of the network (i.e. the bigger it becomes the more resilient it gets), we might stay unconvinced about decentralization as a solution, as soon as we would be facing scalability, decentralization, and security, or downtime issues.

But here comes the twist. Nodes in a decentralized network like Ethernity CLOUDs are location and time agnostic. That makes them surprisingly scalable, resilient, and cost-efficient. Here is how: 

1. Time and location agnostic: This means nodes can stay anywhere in the world, and find the cheapest energy sources (most likely renewables or wasted energy) and the coldest areas, to save electricity bills and cooling costs. 

2. Open-source: Since it’s decentralized and open-source, anyone can participate. Right now you’re easily able to buy pre-configured Ethernity CLOUD Nodes built by official hardware affiliates. You’re also able to offer your computational power to other companies. Soon, after enough adoption, you’ll be even able to set up, build and sell nodes by yourself. This structure enables a global grassroots adoption curve that would be the epitome of scalability in the sense of organic growth.

3. Growth: As soon as the network gets to a certain size it’s becoming more and more resilient toward attacks. This way it turns from downtime risk to uptime strength.

4. Fewer security costs: It’s not only more cost-efficient because node operators will find the most competitive energy source. The decentralized structure allows us to lower security costs massively. First, because there wouldn’t be a single point of failure, and second in the case of Ethernity CLOUD the servers/nodes are hardware encrypted. So even a hacker with access to the machine couldn’t see what data gets processed.

5. No legal cost: Legal costs wouldn’t be needed to be paid because there is no single responsible entity in a decentralized network. The only exception is if some players in the network decide to form a company around these operations and generate an income. Then they would become object of legal responsibilities. Similar to big bitcoin miner companies in the U.S. (e.g. Riot Blockchain).

6. Competition: In general there is an ever going competition. We might even call it a decentralized marketplace for computational power. In this scenario node operators would do anything to keep their expenses down or to specialize their hardware to improve computations by over 10-15% in order to have better prices and more customers in this potential decentralized market. 

Taking all this information into consideration we might compare the cost-effectiveness of on-premise infrastructure, centralized cloud providers, and a decentralized computation network and realize: Some of the costs wouldn’t even occur for a user. Some others would be compensated to a portion by the automated network rewards.

Some costs are still unclear, due to the novelty of such networks. But we can assume that a portion of the “???”-costs will be covered by automated network rewards.

3. Final thoughts

So while cloud computing was historically the logical consequence of the needs of our web industry in the forthgoing into a more digitalized society it had not only moral issues but also economic downsides. For one there is a big security and privacy risk in using a centralized cloud provider, which was described in my first article. For another, we explored the heights of operational costs of these centralized players. So in essence there are two main reasons why we risk that our data becomes a victim of economical interest. Realizing that decentralized structures might not only be technically safer but also more cost-efficient, we finally have a solution that makes privacy protection economical.

This way a decentralized confidential computing platform like Ethernity CLOUDs might mark the path to a web where moral values and economical needs are met at the same time. Not only by technology. Or not only by good wishes. But by the right balance of incentive structures and technical mechanisms, we can enforce this privacy-protecting and cost-efficient environment for cloud computation.

Marcel Ohrenschall