Cloud Computing with virtual machines could be of big value for scientific applications, since it could reduce personnel and cost effort in computer centres. This is because virtualization facilitates computer centres to concentrate on their primary task: providing hardware and computing power. Disturbing software problems leading in crashing machines are displaced to the users, who - in case of virtualization - benefit from the machine image concept that allows to just reincarnate a system from an image file.
It was argued, that the optimal solution for scientific computing would be an own computing cloud for the price of own hardware in combination with an additional and easy accessible commercial cloud, that offers big flexibility itself (like Amazon Web Services do). Easy accessible means that a standard Cloud Computing API should be able to control both computing clouds. This would bring along the huge advantage of being able to switch between clouds in an impressively easy way, resulting in a very convenient solution to balance out peaks of desired computing power.
Motivated by these promising assumptions, I started building a proof-of-principle job system on top of the reliable technical infrastructure of Amazon Web Services (AWS), using the Elastic Computing Cloud EC2 including Elastic Block Stores (EBS), the Simple Storage Service S3 and the SimpleDB. It could be shown that it is possible to move serious scientific computing (ATLAS Computing with an ATLAS Software Release on Scientific Linux 4) to the cloud. Hence, ATLAS Cloud Computing is, in principle, a promising substitute for classical ATLAS Computing in the LHC Computing Grid. This also should apply to any other scientific computing application.
In the future, the next step is the implementation of the concrete plan to rebuild the job system by only using the services EC2 (without EBS) and S3. This step smooths the way in direction of cloud portability, since these two services are the basic Cloud Computing services that should exist as basic versions in any computing cloud. This reduced job system will be available to any other AWS user by using public Amazon Machine Images and storing ATLAS Software Releases centrally on S3 - which will increase usability.
And - to mention the best thing at the end - there already exists an approach to set up an EC2 style cloud with own hardware: http://workspace.globus.org. The great Nimbus CloudKit, developed by the Nimbus people, makes it possible. Of course, Nimbus does not support the whole EC2 API. This is the reason why the AWSAC job system must be reduced to the elementary needed API calls, so that the systems will be able to match each other in the future.
Nimbus raises hope, that it is possible to develop a general Cloud Computing API with different possible clouds at the back end, e.g. a Nimbus Cloud for the price of own hardware and, in addition, Amazon’s EC2 to satisfy peaks of desired computing power.
Cloud Computing is coming and should come in science, too: as recently published (here - sorry, German only), the Steinbuch Centre for Computing - which provides the LHC Tier 1 Centre for Germany - cooperates with Intel, HP and Yahoo to explore the benefits of Cloud Computing for scientific applications, using a huge amount of hardware.
