When you create a web application for the cloud, there are many things that need to be done differently. It's not just a case of saying "I'm doing cloud" and all you're really doing is putting it on someone else's VM. Doing this, the costs are much higher than if the application is designed with cloud in mind. This might be fine from an infrastructure point of view, but the cloud can have profound impacts on development from the ground up.
Azure allows us, and also forces us, to engineer our applications completely differently.
Forces? Well, this is because when you're hosting in an App Service plan, you're billed based on compute time and resource usage. This forces you to write more efficient code.
You can't just write code that performs unnecessary operations and get away with it - the efficiency of code now directly translates to dollars. This demands that you think twice when writing code, especially loops, to ensure you're being efficient. Caching becomes a major priority so that you're not hitting a database unless you absolutely have to.
With the weight of costs, you need to start thinking about more efficient ways of doing everything. Luckily Azure offers many services to help increase application efficiency. The most basic example is storing images and other static resources in Blob storage. This is fast, lightweight, and extremely cheap.
Scripts and CSS can be offloaded to localised CDNs. This also increases speed and significantly decreases cost.
Some data can shift from SQL to the often cheaper NoSQL models such as Table Storage, DocumentDB, or Graph data with its powerful relational features. An application might use all of these data types simultaneously for different purposes. For example, a shopping cart site might use SQL to store a user's identity information, Table storage to manage their shopping cart, DocumentDB to store product details, and Graph data to build "You might also like" links.
Need to run asynchronous/background tasks such as generating thumbnail images? Use Web jobs to run scripts, executable, and third party libraries completely outside of the typical web application Request/Response lifetime. Use queues to decouple logic from the main workflow, or use multiple worker threads to perform parallel processing.
We can also take advantage of Azure Search (powerful search services), Service bus (host web services locally but expose them via the cloud), Azure machine learning (predictive analytics, data mining, and AI), and Notification hubs (push messages down to clients).
Then, we can host our applications in App Service plans (instead of Virtual Machines) and take advantage of Visual Studio Online's build and Continuous Integration features. We can also leverage elastic scaling and Application Insights.
There are many more features in Azure and more appearing every day.
To summarise, developing for the cloud is very different from building a website that will be isolated on a local server. It's a far more distributed model where separation of concerns is much more pronounced than we're used to. Developers and architects will need to think differently when designing applications.