Dimos Raptis - Distributed Systems for practitioners

Distributed systems are becoming ubiquitous in our life nowadays: from how we communicate with our friends to how we make online shopping and many more things. It might be transparent to us sometimes, but many companies are making use of extremely complicated software systems under the hood to satisfy our needs. By using these kind of systems, companies are capable of significant achievements, such as sending our message to a friend who is thousand miles away in a matter of milliseconds, delivering our orders despite outages of whole datacenters or searching the whole Internet by processing more than a million terabytes of data in less than a second. Putting all of this into perspective, it's easy to understand the value that distributed systems bring in the current world and why it's useful for software engineers to be able to understand and make use of distributed systems.

However, as easy and fascinating as it might seem, the area of distributed systems is a rather complicated one with many different execution models and failure modes. As a result, in order for one to simply understand how to use a 3rd party library or verify the correctness of a distributed system under construction, one has to digest a vast amount of information first. Distributed systems have been a really hot academic topic for the last decades and tremendous progress has been achieved, albeit through a large number of papers with one building on top of the previous ones usually. This sets a rather high barrier to entry for newcomers and practitioners that just want to understand the basic building blocks, so that they can be confident in the systems they are building without any aspirations of inventing new algorithms or protocols.

The ultimate goal of this book is to help these people get started with distributed systems in an easy and intuitive way. It was born out of my initiation to the topic, which included a lot of transitions between excitement, confusion and enlightenment.

Of course, it would be infeasible to tackle all the existing problems in the space of distributed computing. So, this book will focus on:

• establishing the basic principles around distributed systems
• explaining what is and what is not possible to achieve
• explaining the basic algorithms and protocols, by giving easy-to-follow examples and diagrams
• explaining the thinking behind some design decisions
• expanding on how these can be used in practice and what are some of the issues that might arise when doing so
• eliminating confusion around some terms (i.e. consistency) and foster thinking about trade-offs when designing distributed systems
• providing plenty of additional resources for people that are willing to invest more time in order to get a deeper understanding of the theoretical parts

Fundamental Concepts

Introduction

What is a distributed system and why we need it

First of all, we need to define what a distributed system is. Multiple, different definitions can be found, but we will use the following:

"A distributed system is a system whose components are located on different networked computers, which communicate and coordinate their actions by passing messages to one another." []

As shown in Figure , this network can either consist of direct connections between the components of the distributed system or there could be more components that form the backbone of the network (if communication is done through the Internet for example). These components can take many forms; they could be servers, routers, web browsers or even mobile devices. In an effort to keep an abstract and generic view, in the context of this book we'll refer to them as nodes, being agnostic to their real form. In some cases, such as when providing a concrete example, it might be useful to escape this generic view and see how things work in real-life. In these cases, we might explain in detail the role of each node in the system.

Figure 1: A distributed system

As we will see later, the 2 parts that were highlighted in the definition above are central to how distributed systems function:

• the various parts that compose a distributed system are located remotely, separated by a network.
• the main mechanism of communication between them is by exchanging messages, using this network that separates them.

---

Now that we have defined what a distributed system is, let's explore its value.

Why do we really need distributed systems ?

Looking at all the complexity that distributed systems introduce, as we will see during this book, that's a valid question. The main benefits of distributed systems come mostly in the following 3 areas:

• performance
• scalability
• availability

Let's explain each one separately. The performance of a single computer has certain limits imposed by physical constraints on the hardware. Not only that, but after a point, improving the hardware of a single computer in order to achieve better performance becomes extremely expensive. As a result, one can achieve the same performance with 2 or more low-spec computers as with a single, high-end computer. So, distributed systems allow us to achieve better performance at a lower cost. Note that better performance can translate to different things depending on the context, such as lower latency per request, higher throughput etc.