The verdict is in: cloud computing should not replace supercomputers for scientific research. Such was the result of a two-year study conduct by the US Department of Energy on the feasibility of cloud computing for meeting the computational demands of big-data research projects.
The 169-page report says that while commercial clouds might be fine for enterprise applications, big data research require more "care and feeding" — in other words, the marketing pitch of the cloud as a plug and play compute solution does not really hold water.
The DOE team, comprised of the Argonne National Laboratory in Illinois and Lawrence Berkeley National Laboratory in California, executed a range of scientific computing projects on Magellan, a testbed for cloud computing with server farms located at the National Energy Research Scientific Computing Center and the Argonne Leadership Computing Facility, as well as commercial clouds such as Amazon's EC2. The performance, costs, and manageability of the clouds were then compared to a Cray XT4 supercomputer and a Dell cluster system.
“Our analysis shows that DOE centers are often three to four times less expensive than typical commercial offerings,” the authors write in the report. “These cost factors include only the basic, standard services provided by commercial cloud computing, and do not take into consideration the additional services such as user support and training that are provided at supercomputing centers today and are essential for scientific users who deal with complex software stacks and require help with optimizing their codes.”
The study reached the following conclusions:
Scientific applications have special requirements that require cloud solutions that are tailored to these needs.
The scientific applications currently best suited for clouds are those with minimal communication and I/O (input/output).
Clouds can require significant programming and system administration support.
Significant gaps and challenges exist in current open-source virtualized cloud software stacks for production science use.
Clouds expose a different risk model, requiring different security practices and policies.
The MapReduce programming model shows promise in addressing scientific needs, but current implementations have gaps and challenges.
Public clouds can be more expensive than in-house large systems. Many of the cost benefits from clouds result from the increased consolidation and higher average utilization.
DOE supercomputing centers already achieve energy efficiency levels comparable to commercial cloud centers.
Cloud is a business model and can be applied at DOE supercomputing centers.
Click here to download the study.