COMPUTING

Introduction

The start of the 2012 run has been busy for Computing. We have reconstructed, archived, and served a larger sample of new data than in 2011, and we are in the process of producing an even larger new sample of simulations at 8 TeV. The running conditions and system performance are largely what was anticipated in the plan, thanks to the hard work and preparation of many people.

Heavy ions

Heavy Ions has been actively analysing data and preparing for conferences. 

Operations Office

Figure 6: Transfers from all sites in the last 90 days

For ICHEP and the Upgrade efforts, we needed to produce and process record amounts of MC samples while supporting the very successful data-taking. This was a large burden, especially on the team members. Nevertheless the last three months were very successful and the total output was phenomenal, thanks to our dedicated site admins who keep the sites operational and the computing project members who spend countless hours nursing the systems. We managed to exceed expectations everywhere, simulating and reconstructing more than two billion MC events since March 2012. Transfers between all sites added up to almost 10 PB in the last 90 days. We expect this level of activity to continue through the busy summer.

Physics Support

Physics Support has identified its key tasks and users have signed up for this years ESP credit in the new structure, which merges the former User Support and Analysis Operations team. Our thanks go to Kati Lassila-Perini for co-leading the User Support project for several years. Kati will now be leading the data preservation implementation effort for CMS, replacing her will not be easy and we thus welcome volunteers in this area. There will be a PAT tutorial at CERN during 2-6 July and CMSDAS school in Asia (CMSDASia) in Taipei, Taiwan during 11-15 September 2012. The workbook is being updated to CMSSW version 5_2_5. On the users’ analysis side, the activity level has been high throughout and keeps increasing as expected due to the approaching summer conferences. Due to the combination of high load, aging tools, and loss of some of the key experienced persons, users continue to experience more problems than we would like, but the system is coping well overall even with the work load and the small support group. We give our best efforts to be adequate for the expected busy summer. 

Figure 7: Number of MC events per month

Computing Integration 

The new workload management system for analysis, CRAB3, based on the WMCore infrastructure used by the organised processing workload management system, is in the last stages of development and entering the integration phase. This system implements new features that will improve reliability, performance, scalability, monitoring and security. The Computing Integration group has deployed a prototype of the system at CERN and is in close collaboration with the developers, providing feedback to help shape the final system.  The Integration group has evaluated the performance of scheduling and executing multi-core workflows, which take advantage of the worker nodes multi-core hardware.  Multi-core processing results in a significant reduction of the memory utilised by the data processing application of about 25% for 8-core nodes (40% for 24 cores) with a small CPU overhead of about 5%.  Following recent recommendations by the WLCG, CMS is also moving towards CVMFS for software distribution to the Grid sites. The experiment software is installed in a central repository and becomes available on demand via a system of caches. In close collaboration between the CMS Computing Operations and Integration groups and CERN-IT the central infrastructure and the necessary monitoring is being set up. The first five CMS Tier-2 sites have switched already to the new method of software distribution. More sites including Tier-1s will follow soon.

Computing Evolution and Upgrade

In the past months the WLCG Technology Evolution Groups (TEGs) completed their work and during the WLCG workshop held in New York in May the guidelines about how the computing infrastructure supporting the LHC experiments should evolve in the next years have been discussed. Among the many interesting items there are two worth mentioning: the implementation of storage federations that will help hide data locality to CMS applications, and the exploitation of multi-core resources for the execution of CMS jobs. CMS is well advanced in both areas and has been driving the decisions of the WLCG TEGs.


by Contributions from I. Fisk, J. Flix, O. Gutsche, M. Klute, S. Belforte, J. Letts, S. Malik, J. Hernandez Calama, G. Grandi , C. Wissing and P. Kreuzer. , Edited by K. Aspola.