Cloud Computing with virtualization could make ATLAS Computing much more reliable and robust. Cloud computing centralises the hardware and decentralises software. It hides complexity from the user of the services. Services are made available in a very flexible way via virtual servers. This results in less administration effort and staff costs in computer centres. So the LHC physicists should find ways to use Cloud Computing to their advantage. Using Amazon Web Services (AWS) is one obvious possibility to determine the possibilities for ATLAS Computing in a cloud.
In summer 2008 Prof. Dr. Thomas Trefzger (Uni Würzburg, Fakultät für Physik und Astronomie) sent me (Jan-Philip Gehrcke) to the Max-Planck-Institut für Physik in Munich, where I stayed for six weeks following Stefan Kluth‘s consideration to use AWS as a platform for ATLAS Computing. The idea was to especially use Amazon’s so-called «Elastic Computing Cloud» (EC2) for simulation, reconstruction and other applications an ATLAS Software Release delivers. We call the whole topic AWSAC: Amazon Web Services for ATLAS Computing.
At first we configured a Linux system that fulfils some special conditions simultaneously: the complete ATLAS Software Release and the Amazon EC2 AMI Tools have to run on it properly and it has to run on the virtual machines of the Elastic Computing Cloud correctly itself, using the «Xenified kernel» Amazon delivers. The way we did it is described in this documentation in every detail.
Then we started developing AWSACtools, providing an ATLAS Computing job system. This system takes advantage of the possibility to start up as many virtual machines in Amazon’s Elastic Computing Cloud as you need. AWSACtools consist of a client- and a server component to manage ATLAS Computing jobs. One job is basically one shell script completely written on your own (you can use the ATLAS Software Release‘s commands in there, of course ). It is executed as root on a virtual machine in the cloud as a sub process of AWSACtools‘ server component. To control jobs’ statuses and to receive jobs’ results with the client component, AWSACtools use other Amazon Web Services like Amazon SimpleDB and Amazon Simple Storage. AWSACtools are documented here in every detail, too.
The proof-of-principle job system shows, that the approach of Cloud Computing for ATLAS can make things really easy. At the end of the documentation the basic services a computing cloud should deliver at least for an ATLAS Computing job system are discussed. It is argued, why a general Cloud Computing API with different possible clouds at the back end is highly desired.
