Speaker
Conclusions
This activity was quite important in order to evaluate the functionalities needed to support the distributed computation of an HEP experiment like SuperB.
The results of the test will surely drive the definition of the computing model of the SuperB, and in order to understand which job and data-management system could be suitable for running in production in the next years.
Moreover this kind of activities will be of interest for others VO that are looking for a tool that provides the fundamental features needed to carry-on a complex computing activity on a grid computing environment like EGI and OSG.
Overview (For the conference guide)
The recently founded Nicola Cabibbo Lab will host the SuperB experiment:
an asymmetric energy e+e- collider and detector that will provide a
uniquely sensitive probe of New Physics in the flavor sector of the
Standard Model.
SuperB distributed computing group performed a detailed evaluation of
DIRAC Distributed Infrastructure in terms of service capabilities,
efficiency and reliability for two main use cases: end user analysis and
Montecarlo simulation production. The new Dirac release 6 has been
configured to respond to SuperB requirements all over the majority of
Dirac functionalities. Data management different setup have been
considered: the native Dirac File Catalogue system against LFC (LHC File
Catalogue) in terms of features, reliability and performance. Test bed
and configuration descriptions has been reported including test and
evaluation results, using sites that belongs to the EGI grid distributed infrastructure together and sites that exploits the OSG middleware.
Description of the Work
The recently founded Nicola Cabibbo Lab will host the SuperB experiment:
an asymmetric energy e+e- collider and detector that will provide a
uniquely sensitive probe of New Physics in the flavor sector of the
Standard Model. Heavy quark and heavy lepton studies requires data
sample of 50 ab-1 and a 1e36 cm-2 s-1 as luminosity target.
DIRAC Distributed Infrastructure evaluation for use in the SuperB
experiment will be reported in this work. The use cases End User
Analysis and Monte Carlo Production have been fine tuned and heavily
tested. We will describe the test bed layout including DIRAC version 5
and new version 6, configuration in terms of sites and sub services. The
efficiency test results will be presented about SuperB end user analysis
and Monte Carlo Simulation workflows. Performance and failure studies
have been measured with different DIRAC setup: in terms of data
management system, a fine grain study has been brought on for DIRAC File
Catalogue system against LFC (LHC File Catalogue) in terms of features,
reliability and performance. A particular care have been placed on
interoperability OSG-EGI Grid flavor evaluation. Moreover the DIRAC
Cloud capabilities have been subject of specific interest, experience in
Cloud general resource exploitation will be presented. The presented
work will include, as conclusion, a comparative analysis among Dirac
system and other submission systems available in the HEP community with
pros and cons of each system.
Impact
DIRAC Distributed Infrastructure evaluation will brought SuperB
community to the choice on adapting such an infrastructure as main actor
in all the major distributed tasks an HEP experiment can be involved. In
fact Dirac propose a complete data management environment covering
several use cases: from mass data transfer as FTS intelligent wrapper to
job data management for stage in and out operations. The workload
management includes a pilot factory and implementation of various policy
regarding the resubmission operation, job brokering and fail over
procedures.
Indeed in this work we will present all the features tested in order to obtain a fail-tolerant solution for the two main distributed use cases that SuperB will fulfill in the future, Montecarlo Production and End user analysis.
In particular we will focus on the capabilities of managing both the input and the output files of both those uses cases, together with the possibility to provide the right data-driven match-making .
We will report also about the test executed on the advanced functionalities provided by the DIRAC File catalogue like the metadata management and the ancestor functionalities.
Starting from the results of this work we can evaluate if the Dirac way of modeling distributed resources could be pushed beyond other solutions as general distributed resource exploitation
framework for small and middle VO community size.
