FOCUS 2000-009 M E M O R A N D U M 26.10.2000 from: Martti Pimia /CMS to: Les Robertson /IT, Hans Hoffmann /CoCoTime, Paul Jeffreys /Focus cc: Marco Cattaneo /Focus, Philippe Defert /IT, Dirk Duellmann /IT, Werner Jank /CMS, Norman McCubbin /Atlas, Helge Meinhard /Chorus, Maya Stavrianakou /Atlas, Gilbert Poulard /Atlas, David Stickland /CMS, Manuel Delfino /IT, Michel Della Negra /CMS, Peter Jenni /Atlas === Following the presentation 'Consolidation of SUN Physics Services' in the FOCUS meeting of 12 October, people from Atlas, Chorus and CMS experiments have discussed the presented proposal together with IT representatives on 19 October 2000. The following summarizes the discussion and the agreements reached between the experiments present. The principle of having two platforms available for validation and verification both of the computing model and of the software is very important for experiments of the complexity and scale at LHC. It is not only for the preparation phase and for the turn-on in 2005, but throughout the LHC experiment 10-15 years lifetime that the use of heterogenous computing facilities must be kept possible. Two different platforms used for development and production ensures that expertise will be better maintained, and standards are enforced. The perspective of an environment based on a single platform at the central integration and coordination point of the software and computing is not acceptable. In the past, transitions from using a single platform have been painful. Today this requirement of two platforms translates to Linux on Intel and Solaris on SUN-Sparc but will likely evolve over the years. It was agreed that the proposed SUNDEV facility could well be a common, shared facility between the interested experiments, implementing the experiment-specific environment, and delivering the allocated share. In addition, partitioning of the system in case of special needs should be provided, for example to run in different OS and compiler versions, or for temporary special capacity needs. Especially, the librarians need a dedicated share of the service to build and test regular releases. For decent turnaround in development, up-to-date hardware should be provided. Possible limitations to the user groups should be done on the basis of experiments involved, not by individuals, as testing will be done by quite many developers, not only by the production teams. The system provided on the SUNDEV facility should correspond to the standard set of software at CERN such that the number of applications possibly excluded should be minimized. Web applications will be needed to configure experiment software, for production management etc. Also sendmail for automatic production follow-up will be needed. Development and quality tools will be needed on SUNDEV to instrument the code, in addition to the tools needed on the SUN desktops used for development. Therefore the standard SUN desktop support should be continued. It was agreed that XDM is not required. The experiments stressed the need to be able to run significant amounts of production on the two platforms. This will help assuring the quality of data, as well as optimising the computing model. Then, for example, a working group in an experiment can run similar amounts of data on both prime platforms. In CMS, there are four working groups on High Level Trigger. At least one of these four groups has to be able to validate results by using two equally strong services on different platforms in the same laboratory, to be really able to make significant comparisons. That would mean a capacity of about 12% of the total capacity available for the second platform. Similar fraction of capacity will be needed in the case of real data analysis once the physics analysis groups start to be active. For efficient production, fast connectivity to data servers is also necessary. The meeting concluded that a high quality service for batch as well as for development are needed for two independent platforms in order to build and maintain the high quality software needed for online and offline event selection, reconstruction and analysis. ===