SLAs can also get complicated. Some providers have different levels of guaranteed uptime for compute and storage -- and refund different percentages of the fee based on the outage. Of course, a refund may be small consolation if the outage takes your critical business service offline. As more and more customers use IaaS for production workloads, the harder it will be for some sort of partial refund to satisfy anyone who has endured an outage or severe slowdown.
All this argues for you to avail yourself of whatever free trials a provider offers. Tweak your configurations, explore and determine the features you need, and calculate and recalculate pricing, especially if you plan to run heavy workloads over the long term. If you get hit by nasty latency episodes or an outage during the trial phase, take note. Claims of consistent performance and multi-data-center redundancy are great, but your experience and those of your peers provide a better gauge.
Compute optionsIaaS providers vary widely in the features they offer. But there's a catch. In fact, you can upload pretty much anything you like to the cloud, which allows providers to almost always say, "Sure, you can do that." But what you really want are prebuilt services, such as the database you need, so you can plug and go without burning endless upload, installation, maintenance, and configuration time.
Perhaps the biggest benefit of IaaS, aside from getting the job done without up-front hardware investment, is the ability to scale when needed. That's why an auto-scaling service -- which spawns more instances automatically in response to increased demand -- is a great feature to have. For obvious reasons, so is a load balancing service.
Many businesses now use IaaS for big, periodic jobs of limited duration. If you want to run some sort of huge actuarial calculation, for example, an HPC cluster with 10Gbit Ethernet connections will reduce the duration of your job and perhaps lead you to consider other high-end jobs you hadn't thought practical before.
Perhaps the sexiest compute service of the moment is Hadoop, a distributed processing framework for processing unstructured data. It's great for analyzing, say, website clickstreams or sensor data in an unlimited number of vertical applications, though you need programmer experts to get the results you want. Some providers host their own Hadoop distributions, while others partner with third parties to run the service. If Hadoop is an important part of your reason to use IaaS, the less Hadoop infrastructure you have to string together yourself to make it work, the better.