Indeed, transparency is a problem. Although I disagree that you need to know where or specifically how the machines are configured. What transparency you do need is how the service is performing. What is throughput? Latency? Response times? Right now, you mostly only know they are on and off - having more of an early warning system is critical to getting your redundancy strategy in place - and more quickly understanding if the problem is really in the cloud or the app.

You pay the cloud service provider to organize the service (and the software that powers it). When it fails, they are responsible for correcting it quickly and paying, or crediting, your account as it exceeds SLAs. You are ultimately responsible for your own SLAs - that means making wise choices about if you can use the cloud at all, and make sure you have an automated redundancy process. Transparency and the tools that can help have been slow to emerge.

Full disclosure, I work for web monitoring provider http://www.hyperic.com/, who is actually working with the major cloud providers to develop a new tool that provides independent transparency to the performance of these clouds - http://www.cloudstatus.com/ and will be integrating that service into the open source Hyperic HQ complete monitoring and mangagement tool we also provide. Its definitely a work in progress, but we believe we're on a strong path to be solving this problem in the near future.