Centralization

What is a Centralized System?

A system is centralized if all ruling power lies with one or a few entities. Participants in this organizational structure must adhere to these rulings and cannot disobey them or make decisions themselves.

The simplest example of a centralized systems is an absolute monarchy. The king or queen is the sole decision maker and does not have to answer to any other authority when making good or bad decisions. Thus, horrible rulers, like in Game of Thrones, may exist who abuse their power for their own twisted agenda.

Another example is oligopolies. In this case, a small group of players in a market hold the vast majority of power and may collude to retain this power and consequently maximize profits. This market dominance allows them to dictate prices and eliminate competition. Again, OPEC is a famous example.

A more down-to-earth example is your typical classroom. The teacher gives their students assignments. If they fail to do them correctly, they fail the class and cannot continue their studies. This system is often not objective; everybody knows the teacher’s favorite and least favorite students. The latter ones are pretty much screwed. They get worse grades even if they are objectively on par with other students. The teacher is the centralized power who can abuse this position and make decisions based on feelings and other unfair measurements.

Centralized Software Systems

Most applications we use every day are also centralized. This centralization is twofold:

• Services are run by a company that holds total control over a system.
• A given software system is built on a centralized architecture.

Owned Systems

Services like Twitter, Facebook, YouTube, Wikipedia, and your online tax office are applications owned and operated by single companies, institutions, or persons. They retain complete control over these services while you are using them. If one of them decides to ban your account from their platform, you can do nothing about this. If they choose to shut down their service from one day to the next, their service just does not exist anymore, and all the data you stored with them for free is also just gone.

You are out of luck as long as they are not forced by laws or contracts to provide their service to you. However, even if they are obliged to serve you, there are still ways to deny you, as they often act as black boxes.

Architecture

Aside from humans controlling software systems, centralized architectures are the predominant style for most applications. A single controlling entity often defines a system’s entire state; all other components must adhere to these decisions. A single point of truth is created.

In a typical client-server architecture, the server component is this controlling entity. It holds and controls the application’s state while clients can read data and might be able to contribute data to the state if allowed to. For example, imagine a PHP web forum that runs on a server. Web browsers, the clients in this case, can read the posts in the forum and add posts if the user has permission.

Another prominent example is distributed database setups with a single writing instance and many read-only replicas. Data changes can only be executed on the writing instance, which acts as the single point of truth in the system. If these changes are accepted, the data is subsequently copied to the read-only replicas. Writing changes to the slave instances is not allowed from external clients but only from the primary node.

Other large cloud systems like Kubernetes clusters are centralized despite being distributed. The network is controlled by a control plane that sets the state for the entire system and thus assigns workloads to the worker nodes. If the control plane fails, the whole cluster becomes inoperable.

As state handling generally becomes complex and annoying relatively fast, state handling in large distributed systems is even worse. Thus, managing the state in a centralized manner reduces the complexity tremendously.

Risks

The two biggest concerns with centralization are the abuse of power and the susceptibility to failure.

As the examples above describe, a single or few ruling instances have total control over a system. This holds the risk that they act maliciously without any further oversight. They might abuse their power for their own gain, or another party could force them to operate in their interest. Threats of violence, orders from their boss, or law enforcement might pressure the humans in control to change a system.

Bugs, hacks, or viruses can compromise a software component. As controlling instances are effectively the single point of truth of systems, they are simultaneously the single point of failure. In the best case, the system comes to a halt; in the worst case, your application misbehaves, creates incorrect outputs, or actively inflicts harm. In either case, human lives are in danger if the given system is used to provide vital services, like emergency services, traffic control, or weapon systems.

Backups, redundancies, and further oversight mitigate these risks to some extent but cannot eliminate them entirely. Centralized systems remain centralized and would otherwise lose the properties the owners and operators wish for.

Users of or participants in such systems must trust the controlling parties as they can do whatever they want: ban you, change your data, and much more.

Are Centralized Systems Always Bad?

No.

It all depends on the environment and the purpose of a system it is supposed to fulfill. For example, if you host your personal website, you probably want to determine what content you are publishing. Likewise, you most likely want the pilot of your airplane to be in complete control as you trust in their capabilities and training while you have no clue about how to fly a plane.

Centralized systems are, in general, easy to set up and understand. They enable fast decision-making and quick action without cumbersome processes. These properties might be essential in certain situations, like for soldiers in the heat of battle or for football players who listen to their coach. But, again, in these cases, we trust in the abilities and the experience of ‘superiors’ or leaders.

For many of us, a centralized system is just natural: Having a leader in your team, having a king, having a boss, and following the instructions of the police or other experts in their field. It is also much easier to trust another entity and offload specific responsibilities on them: The government will take care of X, and your boss will make the right decision for the company. Sometimes those decisions are hard, like the CEO letting people go to save the company. In a more democratic system, such choices might not be possible as nobody wants to lose their job.

But what about a service like Twitter or Instagram? For some people, they are vital infrastructure for communication. However, they are owned and operated by companies for profit. Thus, they want to be in total control of these applications. Users, on the other hand, could not care less about those companies’ profits as the services are free to use for the most part. Due to the enormous amount of users, the applications have become ubiquitous in all aspects of life: communicating with loved ones, reading the news, and getting important information from your government. And now, we trust those companies to handle all our personal data and integrate them as essential parts of our lives. But what happens if they cannot make enough money off of you? Are you willing to pay them for their previously free services? What if you become a liability for them, and they block your access? Imagine sitting on the bus and being unable to browse Instagram; scary.

This shows the conflict of interest we are facing. Should essential systems and services be centralized without the users or participants having any say? Or should certain services simply exist for everyone to use and partake in as a collective undertaking no single entity controls?