A web application like Zoho may all be “in the cloud” as far as users are concerned, but in reality it sits on racks and racks of servers and switches and storage. They occupy space, consume power, and most importantly, they produce lots of heat, which requires even more power to cool them. One other product comes to mind that has the similar kind of power/space challenge – the ubiquitous mobile phone.
Given how mobile phones pack a whole lot of functionality in a tiny package, I have wondered if the ideal server farm is just tens of thousands of mobile phones packed together. It seems to me that the semiconductor technology behind mobile devices is far, far more power efficient than the stuff that goes in servers. Partly it is a backwards compatibility issue, with servers having to run code written all the way back to 1980s, while mobile phones simply didn’t exist that far back. Partly, it is also a function of how traditional client-server applications were architectural monoliths, compared to the deeply distributed “service-oriented architecture” that is common in web applications today.
With mobile phones approaching very respectable CPU & memory capacity, packaging them together as a server cluster makes a lot of sense. Linux can run on almost all of the modern CPUs common in cell-phones, and the mobile version of Java seems actually well-suited for server use, particularly for deeply partitioned, distributed applications. Lightweightness is actually an advantage in server software, just as it is in mobile software.
So the day may not be far off when you use in-the-cloud applications on your mobile device, the same kind of technology powers both your mobile device as well as the servers. It brings a different meaning to the word “convergence”, a different kind of marriage between the web and mobile devices.
On a related note, a company called Azul Systems provides a cluster of hardware-assisted-JVMs in a box. They have specialized semiconductor technology to power Java applications. Such ideas could completely alter the economics of running data centers.