What do cloud services really do? At the risk of oversimplifying, cloud services provide users with pools of shared computer resources to be operated remotely—often by the internet. This section will describe ways in which cloud services can be characterized and typical ways in which data community often uses them.

One very frequent way to characterize clouds is by the type of service they provide:

• Software as a Service (SaaS): Users have access to software through cloud computing. This tends to be less flexible but easier to use and understand.
• Platform as a Service (PaaS): Offers users a platform where they can develop their activities. It is very easy to master, plus it tends to be a little more flexible than SaaS.
• Infrastructure as a Service (IaaS): Usually through virtualization (virtual machines), users are granted access to computational resources. This tends to require a broader knowledge to operate but is far more flexible than any other option.

The best cloud companies available transcend the SaaS, PaaS, and IaaS classification, as they offer enough products for each of those categories. For example, Azure Machine Learning Studio could be well characterized as a PaaS, but, from them, you could also rent many E2s v3 instances, which are better labeled as IaaS.

Cloud providers can also be divided based on their nature—private, public, or hybrid. Speaking about individual cloud products, I tend to split them by what general tasks they are usually allocated for. Basically, data scientists might seek cloud services thinking about the following types of usage:

• Store: People are already used to cloud storage services such as Google Drive, Dropbox, and OneDrive, which are very good for storing personal files. High-level data science may require tons of data to be analyzed, but they have to be stored somewhere first. Storage clouds offer a specialized service that allows users to store huge amounts of data.

• Compute: Compute clouds are usually optimized for RAM memory, GPUs, or both. They are often required to carry out the heavy computations. Computer products tend to work well with storage ones and are a good option if either the local resources can't deal with the computations or there is a need for speed.
• Hosting: Eventually, computations will lead to an algorithm, an application to be deployed as a new service or with an existing one. An option is to rent clouds to host the application online (web hosting), which frequently offers auto-scaling tools that prevent you from being stuck with too few or too many servers.

Depending on how teams are arranged, a data scientist may be called to work only with one of these types or all of them at a time. If they work alone, the likelihood of working in all of them is far greater compared with the folks working in big companies, which may have specialized teams—data curators, hardcore data scientists, web developers—that will only work with only one or two of these types at the most.

Each of these types of clouds comes with challenges of their own. For example, if you are storing data, you might worry about how to compress it to fit in less space while also being easy to recover, query, update, and maintain. While doing computations, you might care about efficiency and accuracy. Folks working with web hosting may be preoccupied with functionality and user experience. All of them should worry about security.

Throughout this chapter, we will explore the Microsoft Azure cloud using minimal examples. Before proceeding on to the practical stuff, let's discuss some important aspects to look for while picking a cloud.