Author(s) |
Philipp Schieferdecker, Vincent Boyer, Angela Brett, Eric Cano, Andrea Carboni, Marek Ciganek, Sergio Cittolin, Dominique Gigi, Frank Glege, Robert Gomez-Reino, Esteban Gutierrez Mlot, Johannes Gutleber, Claude Jacobs, Juan Lopez Perez, Frans Meijers, Emilio Meschi, Roland Moser, Alexander Oh, Luciano Orsini, Lucien Pollet, Attila Racz, Christoph Schwick, Dimitrios Tsirigkas (CERN, Geneva), Hannes Sakulin (CERN, ), Michele Gulmini (CERN, Geneva; INFN/LNL, Legnaro, Padova), Joao Varela (CERN, Geneva; LIP, Lisboa), Samim Erhan (CERN, Geneva; UCLA, Los Angeles, California), Steven Murray, Ichiro Suzuki (Fermilab, Batavia, Illinois), Gaetano Maron, Andrea Petrucci (INFN/LNL, Legnaro, Padova), Jin Cheol Benjamin Kim (KNU-CHEP, Daegu; CERN, Geneva), Gerry Bauer, Christoph Maria Paus, Konstanty Sumorok (MIT, Cambridge, Massachusetts), James Branson, Elliot Lipeles, Marco Pieri, Matteo Sani (UCSD, La Jolla, California) |
Abstract |
The Large Hadron Collider (LHC) at CERN is scheduled to begin operation in 2007 and will deliver proton-proton collisions to the CMS detector at a rate of 40 MHz. A purely hardware based first filter stage, the Level1 trigger, is designed to reduce the rate to about 100 kHz or approximately 100 GB/s. The CMS DAQ employs a second filtering stage to further reduce the rate to about 100 Hz, while the physics signals of interest must be retained. The CMS High Level Trigger (HLT) is designed to achieve this goal by running sophisticated software algorithms on a computing farm comprised of O(1000) CPUs to make a filter decision. Most existing hadron collider experiments use a three-level trigger system, with an intermediate, hardware-based trigger level to reduce the rate to about 1 kHz; CMS relies on a flexible and scalable software-driven HLT to process complete events at the full Level 1 accept rate. This approach is limited only by available computing resources, which can be extended and upgraded without changes to the design. The implementation of the CMS HLT architecture is presented, with focus on the mechanisms to reliably configure, control, and monitor the system. |