Over a million developers have joined DZone.

Scalability Introduction for Software Engineers

DZone's Guide to

Scalability Introduction for Software Engineers

These insights on designing your architecture for scalability will help you prepare for scaling as you are building your systems.

· DevOps Zone
Free Resource

Learn more about how CareerBuilder was able to resolve customer issues 5x faster by using Scalyr, the fastest log management tool on the market. 

Scalability is not only an interesting topic, but it is also a topic where it's really difficult to find well-organized resources for its proper study. It's very difficult, if not impossible, to find the necessary scalability resources teaching you from zero and getting you to be a software engineering expert on the subject. You need to gather your resources. This is why we are very much interested in pushing through and trying to evaluate what scalability is and how to achieve it by finalizing and organizing the topic.

Scalability - It's Not Only About Architecture/Servers

The first question to ask about scalability is whether it is a business problem or a technological problem. Well, you can think about scalability as a mere high-tech problem where you need to scale your databases or services. However, you need to remember that the scalability problem has existed throughout human history. We needed to know how to scale our food growth. We need to know how to scale our car manufacturing. You see, scalability happened before Silicon Valley was filled with programmers.

At your high-tech business, you can view the scale-it challenge as a reflection of real physical processes scaling; these physical processes also need to scale. When we discuss scalability, it is not merely a technological aspect, it is not just a set of diagram or blocks you draw on the whiteboard or just scaling requests into multiple servers.

It's a real-life situation and a problem which you solve in surprisingly similar ways in both low-tech and high-tech industries. When you have an organization, you have people. Some of these people are responsible for creating your architecture, maybe the architects or the senior developers, or, if you are a small company, then all developers are doing architecture! Of course, it's best that also in a big company, all developers are in a position to create their architecture, but still, there is someone or a few architects on top. Scalability is first and foremost a people and business thing to deal with! The people are creating the architecture, so we need to scale our business by enabling our people to create scalable architectures. We first need to enable the people!

Humans Create Architecture

When we are talking about scalability, we should see in front of our eyes people who are creating technology. An organizational structure, as far-fetched as it can be seen in developers' eyes, is actually rather important in that manner. The structure questions, such as what the team size should be, when considered and answered well, would cause you to be able to create better architecture. Now, when you choose your organization structure, you need to take into account a few considerations, and some of them are how much communication you need between people. You want the least amount of communication which will allow people to do a great job.

Obviously, if people need to communicate too much one with another, then not only will your business not grow fast and not be scalable, but also the architecture they create will be less optimally aligned with the solution you want.

Breaking the Rules

Another question to ask is whether you should have rules. A tight rule system has implications. You can choose to have very tight rules and processes, but it can also hinder your ability to scale. That is because if you impose too many rules on the people, they have less space to take off, and when you scale, you need to break some rules. Now, you can say that you apply a rule that enables people to break the rules, but usually this is a snitch, and you need to be able to break your rules in a more natural way. In order to scale, you need to be able to have creative thinking, and creative thinking requires breaking the rules.

CEOs - Software Engineers or Not?

Let's take like the five top IT companies like Amazon, Google, Facebook, Twitter, Microsoft. You see that the CEOs are coming from a technological background, which makes you think that in today's atmosphere, a person with a technological background can create a more scalable solution or ideas. Now, it is indeed very important for anyone to be aware of the architectural decisions your people make- it's super important, but it's not a must. Airbnb's founders are not technical. We will just say here that it's a very high recommendation. It's best, of course, if the CEO is coming from a technological background, but marketing, finance, and other aspects are also super important. We are developers, we tend to disregard these aspects, but they are important and a CEO should be very familiar with them. The claim is this: it's easier for a software engineer to learn finance than the other way around. The CEO should at least understand the core concepts. As an engineer, you tend to disparage them, but marketing and finance are too important to be only possessions of the CEO.

The Beginning of Scaling

When you take any solution or any product you create, the basic question to ask is, how do you scale it when you have multiple ways of doing that process? First, you can duplicate; if you have one person servicing the requests coming for the customers, then you can simply duplicate, or hire more people.

The same goes with the servers; if you have mainly a stateless server, then the basic way in which you scale your solution is to simply create duplicates of this server. Now this is the simplest solution and it has its drawbacks, but it has many advantages. The main advantage is that it is very very simple to implement, but it is also limited because you cannot always do that, you need to design your services first to support that. So the first rule is that if you have a solution or a service or a server that you create, if it is mainly stateless, then you can simply duplicate it. The same goes for a database, which is yet another service, or a file system - yet another service. Whatever you have, simply duplicate if the service was built for it! This would allow for servicing more requests easily.

When you have state, usually if you need to store some state, then if you have a database which is specific for state storage, then you prefer storing the state there and not in the app layer. If you store it in the application server, then you have the problem of which server to access in order to get or store this state. Let's say we have a user named John, and we want to update his name - if we access one server and update the name, and in parallel, let's say that we have another process which updates its address and it accesses a server, then the problem might be that one update in one of my states is conflicting with another update.

You might simply overwrite the other change, so what many solutions do is require you to contact one server to do all the writes, and all writes are going through one server, a so-called "master," and then via replication delegates and replicates. And you read from the slaves. If you don't want to do that, then depending on your specific scenario, you might need to apply some solutions which can do it in a more distributed way, an agreement protocol like Raft or Paxos, but they are more aligned with configurations than real-time data, and there is an agreement overhead for using them.

In this way, you can contact more than one server for an update; as we are able to contact one of a few servers to get the response, we have scaled our system. With agreement protocols, the servers need to agree on the result. They do some processing and communication for that purpose. Before they agree, they ask themselves if they are actually updated with the last value, but what is important to understand from this simple first rule is that we simply duplicate our service, this is rather a trivial right - assuming you designed for it.

Your Money or Horizontal Scaling!

The basic scalability technical issue is all about having your services serve more units of work with more resources. For that, in today's world, we mostly utilize horizontal scaling. We prefer that over vertical scaling. While we will elaborate on that in future posts, let's present the basic plot for you to learn about vertical versus horizontal scaling cost!

The below diagram shows the difference in cost for vertical scalability versus horizontal.

Web Scalability For Startup Engineers
Source: Web Scalability For Startup Engineers.

The author shows that scaling horizontally- for example, with CDN- is not only cost-effective, but often pretty much transparent! “The more traffic you generate, the more you are charged by the provider, but the cost per capacity unit remains constant.”

Choose Simple Solutions

We simply duplicate our service so we can service more requests, and you can do this, as we said, in any component - we simply duplicate the instances of the database - you duplicate the instances of the services. Now, this is very naive, but it's actually good that it's naive, because it forces you to choose simple solutions. As long as you manage to be smart and choose simple solutions which properly answer your problem,  you can apply this to all scaling issues. We will see in further posts that when you cannot apply such a simple rule, or when your scale is too high, then you cannot simply use this method. You will need to apply more methods into your scalability solution, so the first rule is simply to duplicate. If you cannot do that, you will need to apply other scaling methods such as sharding, but let's leave that to the next posts.

We Have Only Scratched the Surface

Hey, we have only scratched the surface here, but this got long enough. I'll get back to the more interesting topics and internals in future posts. The bottom line of this post is that scaling is not only a technical problem; scaling is about whether you can find a simple duplication pattern, about splitting by functionality, about sharding data and services, and last and not least, trying to do it with commodity hardware in a horizontal way!

Find out more about how Scalyr built a proprietary database that does not use text indexing for their log management tool, allowing customers to search 1TB of data in under a second. 

scalability ,software engineering ,architecture ,devops

Published at DZone with permission of Tomer Ben David, DZone MVB. See the original article here.

Opinions expressed by DZone contributors are their own.

{{ parent.title || parent.header.title}}

{{ parent.tldr }}

{{ parent.urlSource.name }}