- Количество слайдов: 30
The Search for Exotic Mesons – The Critical Role of Computing in Hall D
Critical Role for Computing in Hall D The quality of Hall D science depends critically upon the Hall D collaboration’s ability to conduct it’s computing tasks.
The Challenge n n n Minimize the effort required to perform computing Data Intensive Application Compute Intensive Applications Information Intensive Analysis Research Application – methods and algorithms are not fully defined.
Trigger Rates for Hall D Detector 180 kev/s Trigger 15 kev/s 5 k. B/ev 75 MB/s Trigger requires ~100 CPU’s* 5 CPU-ms/ev 100 CPU-ms/ev 1/3 * Full Reconstruction (CLAS) 50 ms/ev today. Full Simulation (CLAS) 1 -3 s/ev today. Assumed detector & accelerator efficiency. Assume a factor of 10 improvement over existing CPU’s
Required Sustained Reconstruction Rate [15 kev/s] * Raw Rate 10 kev/s [1/3] Equipment Duty Factor * *  = 10 kev/s Duplication Factor 5 CPU-ms/ev = 50 CPU’s
Required Sustained Simulation Rate [15 kev/s] Raw Rate 5 kev/s * [1/3] Equipment Duty Factor * *  Systematics Studies 100 CPU-ms/ev * [1/10] = 5 kev/s Good Event Fraction = 500 CPU’s PWA error is determined by one’s knowledge of systematic errors. This requires extensive simulations, but not all events simulated are accepted events.
Annual Date Rate to Archive Raw Data 75 MB/sec * (3 *107 s/yr) * (1/3) = 0. 75 PB/yr Simulation Data 25 MB/sec * (3 *107 s/yr) = 0. 75 PB/yr Reconstructed Data 50 MB/sec * (3 *107 s/yr) = 1. 50 PB/yr Total Rate to Archive ~ 3 PB/yr
Initial Estimate of Software Tasks & Timeline
Critical Software Issues n Early creation of a “core group” of software developer. n n n Key Software Problems. n n n Creation of key design elements. Commitment to key design goals. Software organization and management. Data formats for raw and derived data. Software for defining and accessing raw and derived data. Event Visualization. Using available software. Developing & maintaining high-quality software.
Collaboration Computing Organization n n A single large, common, democratic computer facility will make it impossible for us to complete our computing. Alternative: Provide several sites with the resources required to complete specific tasks. Choose those sites which are lead institutions in specific efforts, such as simulations, calibrations and partial wave analysis.
Collaboration Computing Organization n n Clearly establishes responsibility for software subsystems. Gives University groups working on software something to show for their efforts. n n Can leverage other University and National resources. n n n Helps to attract people and resources to the computing efforts. Infrastructure, personnel, funding, NSF & DOE ITR initiatives. Eases the creation of customized (Grid) computing systems. Establishes new capabilities within the JLab/NP community. n These capabilities allow JLab to take advantage of new opportunities.
Collaboration Computing Organization n The job is too big to be managed without databases. n n n Provides wider access to experimental information. Databases are optimized for managing large data sets. We will create 5 – 10 M files every year. Database use can be organized to minimize it’s impact on time critical applications.
Collaboration Computing Organization n Attracting physicists to work on software is difficult. n Perceived importance is based on capital “$’s” spent. n n n Accelerator Detector Computing. Once it works, they have nothing they can show to their dean and say, “I built that!” “Everyone” thinks it is easy. One good way to have a really positive impact on the science. Helps train and attract students for a variety of careers.
Computing Organization Issues n Recommendations. n n n n Online database – rely totally on automated methods. Offline database – rely totally on automated methods. Integrated online/offline/simulation database. Event Analysis – do it at Jefferson Lab. Calibrations – possible to do elsewhere. Physics Analysis – possible to do elsewhere. Simulations – possible to do elsewhere. PWA – possible to do elsewhere.
Computing Organization Issues (continued) n Recommendations. n n Develop infrastructure to easily share computing resources and information. Develop customized computing approach to Hall D computing. Provides clear lines of responsibility for software and computing tasks. These are social decisions – not technical or financial decisions.
Online & Offline Analysis n Integrated online & offline analysis systems. n Pros: n n Common system requires less effort. Encourages cooperation between online & offline. Potentially higher reliability. Challenges: n n Broad contributions to offline analysis require standards and convenience performance overhead. Level 3 trigger performance must be acceptable. No working Level 3 trigger system at JLab. No “suitable” memory management system for CODA events.
Online, Offline & Simulation Database n Automated Experiments Database. n Pros: n n n Common system requires less effort. Encourages cooperation between different computing groups. Better organization of needed information. Higher reliability and better access. Challenges: n n Anyone software developer in the information chain can break it. Distributed simulations require modern organization of the database.
Where to Perform 1 st Pass Analysis? n 1 st Pass Analysis at JLab. n Pros: n n Don’t need to transport the data. Computer system support is in place. Detector experts on site. Challenges: n n Oversubscribed computer system. Obtaining efficient tape access, system throughput is unlikely in a heterogeneous computing environment.
Where to Perform Physics Analysis? n Physics analysis is done where the researcher live. n Pros: n n n Not competing with major analysis & simulation efforts. Easier to involve more people. Challenges: n n Requires a portable analysis code. Requires a good system for quality control of results.
Where to Perform Simulations? n Simulations done at a few institutions. n Pros: Get more groups invested in simulation effort. n Probably don’t need to transport the data. n Easy to do remotely. n n Challenges: Need computer infrastructure in place. n Need software infrastructure in place. n
Experiments Database 1/M Run Detector Config. 1/M M/M 1/M Analysis M/M Calibration M/M Simulation
Related Computing Trends n We depend on commodity computing n n Intel’s Merced processors (Itainium) n n n Clusters Networks Storage Media (disks & tapes) 500 MHz, 64 bits, 4 -way processor A year late File Size n n n Currently 2 GB software limit 2 GB going to 232 * 2 GB (effectively infinite for us) What determines the optimum file size?
Related Computing Trends (Continued) n Grid Computing n n XML – not just a better HTML n n n High speed networks Distributed “service” or “data” centers GLOBUS, Legion, home-grown Standard method for creating self-describing data Many tools available (B 2 B) Mobile Computing, Portal Technology n n Customized access to computing resources via data starved devices Customized view of an experiment or equipment
Key Differences Between Halls B and D n n n More uniform physics goals in Hall D. Jefferson Lab computing infrastructure is in place. Hall B computing personnel hired late in the process. n n Fundamentally changed the direction of the software and organizational approach to the problems. Many things had to wait until the very last minute.
Focus Accurate, Timely Analysis n n n Provide people with the information they need to conduct their analysis Provide it reliably Provide it in the way they need it Provide it efficiently (speed, effort) Provide flexibility for other applications
Efficient Information Access as the Foundation of the Computing Philosophy Simulations Data Acquisition Raw Data, Experimental Conditions Calibrations Data Reduction Hall D Experimental Information Physics Analysis PWA Information From Researchers
Benefits of XML n Standardized access to databases and applications. XML Application XML to DB Config. View DB XML App DB to XML Select XML to XML Select Launcher
Benefits of XML n Standard routines exist in Perl, C++ and Java for converting between internal and external storage. XML App XML SII App SII SII
Obtaining Optimum System Performance a 1 a 2 a 3 a 4 a 5 b 1 b 2 b 3 b 4 b 5 c 1 c 2 c 3 c 4 d 1 d 2 d 3 d 4 e 1 e 2 e 3 e 4 f 1 f 2 f 3 f 4 g 1 a 5 b 4 c 3 d 3 e 3 a 2 b 1 b 5 c 4 d 4 e 4 a 3 b 2 c 1 d 1 e 1 f 1 a 4 b 3 c 2 d 2 e 2 f 2 e 4 g 2 f 3 f 4 Equilibrium Start-up w 3 w 4 x 1 x 2 x 3 x 4 y 1 y 2 y 3 y 4 z 1 z 2 z 3 z 4 x 3 y 3 z 3 w 4 x 4 y 4 z 4 x 1 y 1 z 1 x 2 y 2 z 2 Shut-down 2 Tape Drives 4/1 ratio of processing to I/O per tape 1. 2 TBytes of Disk Required
Estimated System Efficiency