CDF Grid Fermilab费米实验室的CDF网格.ppt

CDF Grid Ian Fisk Igor Sfiligoi On-Going Work CDF CAF and distributed DCAF’s are a significant resource 10k jobs per day from 100 users 95% Efficiency, 200-600MB/s IO ~1/2000 failure rate 5000 GHz CPU equivalent over 10 sites Similar in capacity to Grid3. Propriety Elements Relies on Kerberos for authentication FCP for transfers CAF Monitoring and accounting Current Status CAF Infrastructure has a number of important capabilities that should be maintained for user analysis Interactive elements in a distributed batch queue. Ability to investigate the output of a running process Ability to debug a running process All improve the user’s ability to determine what is happening At the same time a lot of predictable computing may not need the interactive elements Reprocessing can be run in ordinary batch Monte Carlo generation, both centralized and user level, is robust and predictable enough to assume a completely batch model Large grid projects LCG, OSG, and NorduGrid have all been most successful at predictable computing. Goals Allow CDF CAF/DCAF infrastructure to use more common components and have ability for better support while maintaining the interactive components for user analysis Permit opportunistic computing for applications like Monte Carlo generation that don’t need interactive services Allow scheduled shared use of resources for repetitive and predictable computing, like reprocessing Short Term Goals (1) Many DCAFs are running the FBSNG Queue and corresponding service implementation Plan to encourage all current FBS DCAF sites to migrate to the Condor CAF at the time of the DCAF workshop Will be more supportable to consolidate Allows an easy transition to enable those physical resources with other grid interfaces (LCG or OSG) Provides the potential of sharing larger Condor installations available at some of the current installations Work will be performed by local DCAF admins. Elliot Lipeles and Shih-Chieh Hsu, CAF support from UCSD, will provide in