A virtual machine does not an application make; cloud applications need an array of services to function optimally. Comparing unit prices to assess changes in dynamics and competition is fraught with issues – just because VMs are becoming cheaper doesn't mean that applications, the real consumed entity in the cloud, are becoming cheaper, too. Other line items can quickly inflate the price tag.
The CPI now provides a basis for assessing changes in cloud pricing based on how an application is used, not just on point services. To evaluate a market as broad as cloud based purely on the prices of virtual machines and storage is foolhardy. The typical cloud provider doesn't just offer these basic services – it offers a range of services, including database, compute and SaaS. Each of these services has its own differentiated features, value propositions and pricing. Perhaps more importantly, the combination of services and their integration varies from provider to provider. So, although virtual machines are perhaps becoming more price-sensitive and less differentiated, that is not necessarily the case for all cloud services and all cloud providers. This is exactly why the CPI adopts a basket-of-goods approach. In a similar manner to inflation, which is based on a selection of groceries that a typical consumer might purchase each week, the CPI is essentially a number of baskets of goods that define the services needed to operate a number of typical applications.
Basket of Goods: Large Web Application
By defining standard specifications of typical applications, it's possible to measure changes in the typical price of real-life scenarios. The CPI should be used to provide a benchmark to aid planning and decision-making. Enterprises, service providers, vendors, investors and partners should use the CPI, alongside other data and experience, to assess the cost and value of the IaaS, PaaS and SaaS they offer and consume. Value is the key term here – cheaper is not necessarily better. More expensive solutions are likely to be better value if features, support and geography justify the additional expense.
To calculate the CPI, we use a weighted average based on market share. To capture the total cost of hosting our specifications from a range of providers, we seek quotations and derive estimates using online pricing. Providers are asked to submit quotations that deliver as close as possible a match to our required specification, using their cheapest US datacenter. Each total cost is multiplied by that provider's market-share percentage to produce a provider-weighted cost. Market share data is obtained from 451 Research’s Market Monitor service. The weighted costs are then added together and divided by the total market share represented by all providers. In other words:
where for provider i of n providers, si is market share ratio and pi is the market price.
The equation gives us a value that is an approximate representation of the average price an end user would pay per hour for running that application across that percentage of the market. It is not possible to achieve 100% coverage of the market because providers will leave and enter the market, and many providers choose not to disclose their pricing. However, because the prices of our baskets are based on the largest providers able to contribute, it is highly likely those not included are outliers in the landscape with a relatively small impact. Furthermore, many providers surveyed were unable to deliver to our specification (particularly with regard to our large Web application).