Canon et al. give what I think is the most realistic assessment of the value of cloud computing to scientific applications. They point out that the cloud is at the peak of inflated expectations in the Hype Cycle, and set out to get it on the path to the Plateau of Productivity:
- Clouds are simple to use and don’t require system administrators.
- My job will run immediately in the cloud.
- Clouds are more efficient.
- Clouds allow you to ride Moore’s Law without additional investment.
- Commercial Clouds are much cheaper than operating your own system.
They conclude that
- “Cloud Computing as it exist today is not ready for High Performance Computing because
- Large overheads to convert to Cloud environments
- Virtual instances under perform bare-metal systems and
- The cloud is less cost-effective than most large centers”
Commercial clouds are however effective in the following examples:
- Individual projects with high-burst needs:
- Avoid paying for idle hardware
- Access to larger scale resources (elasticity)
- High-Throughput Applications with modest data needs:
- Monte-Carlo simulations
- Infrastructure Challenged Sites where facilities cost much greater than IT costs
- Undetermined or Volatile Needs: Use Clouds to baseline requirements and build application in-house