Lately there have been several attempts to pit on cloud against another using sets of sophisticated (and sometimes not so sophisticated) monitoring tactics. A more recent effort on the part of Bitcurrent and Webmetrics took Amazon, Rackspace, Google, Terrermark, and Salesforce to task on their ability to perform with a range of application types. Part of the reason why this benchmarking effort is interesting is because two of these are PaaS offerings (Google and Salesforce) while the others are IaaS providers.
The cloud performance test involved four types of evaluations to benchmark five native applications and watched as the platforms requested a small object, a large one, performed a CPU-intensive task, followed by an I/O intensive task.
The authors came to several interesting conclusions, two of the most compelling of which are the fact that with a PaaS provider, when the cloud starts slowing down, the whole thing slows down—for everyone involved. When using an IaaS provider, there’s division of the CPU and server’s responsiveness but “you are still contending for shared storage and network bandwidth.” They also found that for those making a decision between PaaS and IaaS, the final choice is dependent on the type of workload . as Abel Avram writes, “if you’re willing to re-code your application to take advantage of big data systems you can scale well by choosing a PaaS cloud. On the other hand, if you need individual machines, you’ll have to build elasticity into your IaaS configuration.”