Today, as Innokrea, we will talk a bit about what a distributed system is, what its properties are, and we will also see an example architecture of such a system. If you're interested, we invite you to read on!


What is a distributed system?

There are various definitions of what a distributed system actually is, ranging from formal and scientific to very practical ones. Some definitions indicate that a distributed system consists of many components or programs located on multiple nodes or computers. Others, however, not only emphasize distribution but also, for example, that a distributed system should appear to the user as a coherent system, and that there are various planes of distribution. However, certain attributes can be identified that characterize such software types, such as:

  • ability to operate on multiple computers,
  • high scalability,
  • heterogeneous technological stack,
  • concurrent execution of multiple processes,
  • failure of a single component does not cause the system to stop functioning,
  • geographical distribution,
  • data replication on multiple computers,
  • need for coordination between different processes to perform operations,
  • asynchronous data exchange,
  • use of networks for data exchange.

Of course, we haven't listed all the characteristics here, but the above ones can give you some intuition about such systems. It should also be noted that not every distributed system fulfills all these characteristics.


Planes of Distribution

The three most important planes on which applications can be distributed are: processing, control, and data. Depending on our requirements, we can manipulate the degree of distribution in these aspects. Processing distribution means that there are different computers connected to a network processing certain data, such as information retrieval or matrix multiplication. Nodes receive part of a certain task and expedite its execution if parallel data processing is possible. As for data, we can talk about their replication or so-called 'data partitioning,' which means storing them in different locations. An example here is databases, where problems with their replication and synchronization are properly resolved by database engines. As for control distribution, it involves transferring some control to another computer node. Examples could be a P2P network or a router, which makes routing decisions based on local information. Control over message passing is thus distributed. Let's try to analyze the client-server architecture in terms of the above criteria:

  • processing - the distribution is not significant because the client typically processes UI information or input data from forms. The server side handles user requests processing or communication with the database. Depending on the type of application, the processing proportions in this architecture may look different (client-side rendering vs server-side rendering).
  • control - on the client side, responsibility lies in initiating requests and handling responses. The server controls business logic or data access.
  • data - data are entirely stored on the server side, so there is no distribution here unless the database is replicated or 'data-partitioning' occurs among multiple databases.

Therefore, it can be said that the client-server architecture itself is neither a particularly distributed type of architecture nor centralized. However, this analysis could be heavily modified for certain scenarios, which wouldn't be unusual. Here, we provide a general outline of such a system, and how distributed it should be to make the system optimal remains at the discretion of the designer.


What is System Transparency?

One of the most important characteristics that a distributed system should have is the best possible concealment of the fact that it operates on multiple nodes. Thanks to this, the system's user can use it without worrying about technical details as if it were operating on a single station. There are different types of transparency, including:

  • access - hide differences in access to added and applied formats,
  • location - hide where the object is located,
  • concurrency - hide that multiple users can use the object,
  • error - hide errors and their correction,
  • migration - hide that the object is moved to another location,
  • relocation - hide that the object is moved to another location while being used simultaneously,
  • replication - hide that the object is replicated in another location.


Where Are Distributed Systems Used?

Both in the academic and commercial worlds, various types of distribution are used due to much greater possibilities of horizontal scaling (where we increase the number of stations to achieve greater computing power). Vertical scaling has its serious limitations and becomes uneconomical above a certain level. Clusters and supercomputers are used for computationally intensive tasks such as language model training or weather modeling. There are special platforms for distributing tasks to multiple computing nodes like Apache Hadoop.

In this regard, we recommend a series of articles about supercomputers:

As for data, any solutions for storing large amounts of data can be mentioned, such as Google's Bigtable or Amazon's DynamoDB. Data are distributed among multiple nodes, allowing for the storage of vast amounts of data for later processing or directly saving information generated from processing other data. Control distribution, on the other hand, is used in handling energy or environmental systems, where software components can make local decisions without the involvement of a central server.

Classic examples of distributed systems familiar to programmers also include microservices used by companies like Netflix.


Netflix Architecture Diagram

Figure 1 - Netflix Architecture Diagram [1]



Distributed systems are ubiquitous today, and we often use them without realizing that they are composed of many software components. The transparency of solutions contributes to user experiences, allowing them to benefit from fast and often reliable systems.