Moving to the cloud

The cloud computing has developed rapidly during last few years. As shown by Gartner study, the cloud market has grown by 18% in 2013 and is now worth 131 billion (or thousand million) US dollars. By the year 2015 it is expected to hit the value of 180 billion. Despite some well known security and privacy concerns regarding public cloud storage, and even despite the PRISM global scandal, the future in which most (if not all) of our data is stored in the cloud seems inevitable. This requires a substantial shift in approach to software architecture and creates new challenge for software developers.

First, you need to abandon huge, complicated, monolytic applications in favour of platforms built from many small, interconnected services. Such services require much less resources, and can be easily replicated througout the cloud, increasing the overall stability and safety of the applications. One of the most famous examples of a successful SOA revolution is Amazon, which radically changed its software platform.

Second, you need to change the underlying technologies. For example, Java with its extremely resource hungry virtual machine, bloated frameworks and heavy threads seems a rather bad choice. On the other hand, Node.js with its small memory footprint and single-threaded, event-driven, non-blocking I/O model fits the cloud perfectly.
If you have ever had doubts about Javascript being the right choice for an enterprise application, than you should read about Paypal moving their backend from Java to Javascript. Except from huge cost savings (Paypal has been extensively using very expensive Java soultions like Terracotta BigMemory to scale its backend vertically) and increased productivity (engineers claim to be able to develop software twice as fast as in Java), Paypal also benefits from new skills of its software developers, who can now work both on frontend and backend using the same programming language.

Third, you may need to redesign the application flow and algorithms used. You may often get better results processing huge amount of data in small chunks on many small cloud instances, than all at once on few huge mainframes. It means not only using MapReduce to speed up your tasks, but also avoiding of using complex sequential algorithms. For example, in typical situations mergesort may perform worse than heapsort, but is much simpler to parallelize and with enough cloud instances will sort your data much faster.
But there is more to it than that. AMD has just announced a new line of 64-bit ARM server CPUs, which is widely regarded as the end of its race against Intel in high performance computing. ARM processors are less complicated and more power efficient than their Intel counterparts, but are also slower - which means that they will be more suited for clouds with many small instances.

Finally, there is a whole new class of problems in cloud computing, which are either not present, or long solved in traditional applications, like data consistence or dealing with transactions. Some of them cannot be solved by software, and they require change in your business model: for example you may have to trade real time consistency for the amount of traffic your application can handle. Also, most cloud installations suffer from poor I/O performance, because there are usually many virtual instances simultaneously trying to access one physical device like hard drive or network interface.