Cloud computing is a very useful if you are processing large amount of data and cannot wait for your results or these are clinical and mission critical operations. However, there are several steps you need to understand before using it for imaging.
The factors that affect most calculations for cost and use are:
RAM: Random access memory. How much RAM do you need for your server. Typical values range from 512 Mb to 64G.
Storage: How much application and data storage do you need. This varies from 1Gigabyte to 1 Terabyte.
CPU: Computing power Faster CPU (Computing processing unit) obviously cost much more.
Operating system: Here there are primarily two choices: Linux vs. Windows
Moving data: There are finite costs associated with moving data to the cloud and then moving it back after they have been computed. These tend to vary depending on the environment but could also vary depending on whether the data is going to be stored in the cloud or just moved there temporarily.
For example, to rent a Windows on-demand server for about an hour of computing costs just 10c in NE America but larger on-demand servers can cost about 90c per hour. However, the real costs may be significant. Consider a Amazon standard small rental with 10Gb, 1.7Gb RAM, 10GB Hard drive, 1 Virtual CPU and 1 ECU (The ECU is defined as the compute capacity of a 2007 Xeon processor) works out to be about $45 which includes data transfer charges. http://www.cloudorado.com/
A recent publication (Nature, 10(2), p96, Feb 2013) highlights the usage of this cloud computing technology using the technology of Bayesian localization microscopy to convert a standard resolution microscopic image to a single molecule high resolution image. However, there is a significant computation cost associated with the resolution. For example a 50 x 50 pixel computation takes 1 full day! The authors have used the Amazon EC2 cloud and packaged the executable program and the running environment into a Amazon machine image that can be readily run by other biologists. This enabled the computation of nearly 12 times larger image in the same amount of time that it would take a 3.5 Ghz Intel core i7 Desktop. They estimated the price to be $25 to process a 25micron by 25 micron image.
Obviously the costs that are associated with this computation are termed in accounting as consumable cost as opposed to the fixed/capital costs of buying a computer system. However, it does make the computation feasible in many cases that were not possible before. In the future, large data analysis will require computation of large data sets that can be very useful moving forward.