We won’t go into the actual costs at this point as it can be complicated and is always getting reviewed. However it’s well worth understanding the basics so that any application you design takes this into consideration. The savings can be significant when compared to hosting applications locally especially when these applications are mission critical.
 

Each Web Role and a Worker Role is a virtual machines (small, medium or large) and will cost you once instantiated even if they are not being used (so be aware of inactive roles including those in the QA environment) . Another common mistake is to generate a worker role for each background task, this can prove to be costly, it is always cheaper to use a technique like Parallel Programming within a single role to minimise costs.
 

The ‘No SQL’ storage is incredibly cheap and can store gigabytes of data for cents/pennies. The SQL storage is by comparison very expensive, but is by no means redundant as it’s query capabilities still outshine the ‘No SQL’ storage. Using the two storage mechanisms together seems to be the best solution. For example it is foolish to store BLOBs in SQL when ‘No SQL’ can do the same for less.
 

Transaction costs are the costs of moving the data outside of the Azure Data Centre and can affect how we build a system. To minimise the costs try to ensure all queries go via an Azure Role, as this will minimise costs. Frequently sending large chucks of data directly to end users will incur high charges.
 

There are costs for using the Service Bus, and the Access Control. However these costs are very small and we found they didn’t contribute significantly to the overall cost. As for the CDN and the Cache costs, we haven’t used these services and cannot comment on the costs.