Can Clouds Replace Grids? A Real-Life Exabyte-Scale Test-Case

The world’s largest scientific machine – comprising dual 27km circular proton accelerators cooled to 1.9oK and located some 100m underground – currently relies on major production Grid infrastructures for the offline computing needs of the 4 main experiments that will take data at this facility. After many years of sometimes difficult preparation the computing service has been declared â€ﾜopen” and ready to meet the challenges that will come shortly when the machine restarts in 2009. But the service is not without its problems: reliability – as seen by the experiments, as opposed to that measured by the official tools – still needs to be significantly improved. Prolonged downtimes or degradations of major services or even complete sites are still too common and the operational and coordination effort to keep the overall service running is probably not sustainable at this level. Recently â€ﾜCloud Computing” – in terms of pay-per-use fabric provisioning – has emerged as a potentially viable alternative but with rather different strengths and no doubt weaknesses too. Based on the concrete needs of the LHC experiments – where the total data volume that will be acquired over the full lifetime of the project, including the additional data copies that are required by the Computing Models of the experiments, approaches 1 Exabyte – we analyze the pros and cons of Grids versus Clouds. This analysis covers not only t echnical issues – such as those related to demanding database and data management needs – but also sociological aspects, which cannot be ignored, neither in terms of funding nor in the wider context of the essential but often overlooked role of science in society, education and economy.