The idea is pretty simple. With traditional hosting one can rent a single box and run dozens of small web sites. With Azure you are expected to rent one worker role per site, which can dramatically increase your hosting costs. The minimum cost for one worker role for a year is approximately $1,051.20 plus storage, bandwidth, and transaction fees. This is why people are asking for a way to make Azure cheaper for very small services and web sites.
The counter-point to this argument is that Azure really isn’t meant for small sites. Underlying this theory is that if you don’t need multiple website roles to host a given site, you really aren’t part of Azure’s target audience. Of course this makes it hard for small sites that may grow into large sites at some point in the future.
This discussion has prompted Jouni Heikniemi to propose creating “Windows Azure Express” and Windows Azure Compute Small Business Edition”. The concept is simple, why pay for an entire processor core when a well written website only needs a fraction of that power. Instead, he would like Microsoft to support plans that would allow developers to rent parts of a compute instance for a target price of $10/month.
Next up is continuing to offer developers free access. Without costly MSDN subscriptions, independent developers who want to experiment with Azure have no choice but to shell out their own money. There isn’t much debate on this topic, but ensuring that developers don’t use it as a way to get free production instances is hard. Perhaps offering something like Jouni Heikniem’s idea would solve this issue as well.
Number three is rather surprising. Azure doesn’t support sending emails, an essentially component of most websites. Ideas include charging per email, placing a hard limit on outgoing emails, or integrating with other email servers like Exchange Online or Gmail.
The next two we are going to discuss deal mostly with the needs of worker roles. Worker roles have two fundamental problems. First, the Azure architecture is designed around the idea that each worker role does one specific task. If you bundle multiple functions together, upgrading the role becomes problematic because you have to upgrade the entire role atomically. The Azure pricing model is the exact opposite. Since you are charged per deployed instance, you are better off combining as many processes together as possible, especially if the processes are usually idle. For this reason there is a suggestion that multiple roles should be able to share the same compute instance.
The second problem is that there is no way to schedule workers roles from within Azure so that they only run when needed. A project could easily need many worker roles that only run at set times and, once completed, should shut down automatically. Since the Azure pricing model is based on how long a role is deployed, not how long it is actually running, using internal timers isn’t an option. External timers hosted outside Azure could be used, but this leads to reliability issues.