This posting is a summary of Lesson 3 in the Cloud 101 Class, in which different methods of allocating cloud computing resources to meet customer demand were discussed, and how those methods affected costs.  You can return to the first posting in Lesson 3 here.

• There are three major ways that vendors allow customers to request resources from the cloud: fixed instance size allocation, quantized instance size allocation, and variable instance size allocation. Each method allows customers to request computing resources according to their needs, which potentially offers significant cost savings compared to purchasing or leasing computer hardware.
• Some vendors use a fourth method in which instance size varies based on the number they pack onto their servers, which does not offer the guaranteed resources that people expect from the cloud, though it is often offered at a fixed price. This is more of a 'hosting' offering than true cloud computing.
• Fixed instance allocation such as that used by Amazon EC2 offers low instance cost, but encourages customers to over-buy instance sizes to avoid performance problems, which increases the actual cost to the customer. Responding to changes in demand requires allocating new instances, which is complex and costly. Third party vendors have arisen to solve some of these problems.
• Quantized instance allocation solves some of the problems with fixed instance allocation by giving customers more flexibility in selecting instance sizes.
• Variable instance allocation allows customers to request exactly the resources they need from the cloud, but requires them to know exactly what those requirements are. Most customers are not familiar with the techniques necessary to understand their true requirements.
• Each allocation method requires customers to understand how much resources their applications need at a far more detailed level than ever before, potentially creating the risk of performance problems or unexpected charges.
• Automatic scaling, either by adding instances or by resizing existing ones, can simplify the requirement of understanding how much resources an application needs to run successfully.  However, it can result in unpredictable bills.
• Because hardware is always changing, cloud vendors may not be able to guarantee an exact performance level per unit of cost. How serious this problem is depends on the policies of the vendor, which may not be transparent to the customer.
• Cloud vendors have to keep unused inventory of servers available for peak demand, which can increase costs. The exact amount of these costs is not generally known.
• Total Cost of Deployment, the actual cost the customer must pay to run their application, is difficult to calculate for cloud deployments (though simpler than TCD calculations for building your own datacenter.) This difficulty threatens the success of cloud computing, since customers are looking a simple way to understand their costs, not just find the lowest costs.