Public clouds for scientific applications cost up to 13x more


Public cloud computing environments for scientific applications have the potential to be up to 13 times more expensive than existing Energy Department supercomputing centers, finds a final report of a 2-year federal study.

The study (.pdf), released in December, chronicles the build out of a private cloud dubbed Magellan by the Argonne Leadership Computing Facility and the National Energy Research Scientific Computing Center, as well as a study of attitudes to cloud computing among scientists and cost estimates of running scientific applications in a commercial cloud.

In that last matter, the study find that moving NERSC or ALCF into Amazon Web Services offering would cost many times more than their total annual operating budgets. In the case of NERSC, the study estimates total cost in cloud to be $200.39 million yearly--potentially a difficult bill for an organization that has only $55 million a year to spend. For ALCF, the cost would be less; just $179.2 million, though its total annual budget is also a correspondingly lower $41 million.

As for how Magellan fared, the report says researchers encountered "performance, reliability, and scalability challenges," although it adds that the open source cloud software stacks utilized to implement the private cloud significantly improved over the 2-year course of the study.

Nonetheless, the software stacks have gaps when addressing many DOE security, accounting and allocation policy requirements, the report adds.

"Even though software stacks have matured, the gaps and challenges will need to be addressed for production science use," it says.

Energy can nonetheless learn from cloud computing, the report says. A survey of NERSC and other federal supercomputer users show that 59 percent said they'd like to control software environments and 52 percent said they'd like to share their software or experiment set up with collaborators--tasks both difficult to do with today's supercomputers. (Data from the survey also suggest that what scientists really care about when it comes to computing environments is having more, since the largest percentage of respondents, 72 percent, said an attractive feature of cloud computing would be "access to additional resources.")

Report authors recommend that supercomputer resource providers investigate mechanisms to provide on-demand resources. They should also consider mechanisms to provide scientists with tailored environments, they add.

For more:
- download the Magellan Final Report (.pdf)

Related Articles:
Exaflop supercomputer receives full funding from Senate appropriators 
GAO reproaches NNSA for nuclear simulation supercomputer disaster recovery plans 
NNSA turns to computer modeling to plug infrastructure data gap