4. WBS 2.2 Software Manager’s Report (Torre Wenaus, BNL)
Much of the work in this reporting period was focused towards the major milestone of the so-called "Lund release" (2.0.0), scheduled for early July 2001 to support physics analysis work leading up to the ATLAS Physics Workshop in Lund in September. This release has the following deliverable set for Framework and related software:
These were all available at the end of the reporting period. Other significant developments within the US effort are outlined below.
The bulk of U.S. ATLAS database development effort in this reporting period was directed toward support of a successful major release of ATLAS software, the 2.0.0 “Lund” release. Two major components of this work are the database infrastructure needed to support the ATLAS fast simulation program (Atlfast) in Objectivity and ROOT, and support for inter-object references.
It was not anticipated at the beginning of the quarter that database support for Atlfast would occupy such a significant fraction of core database developers' time. At the request of the ATLAS Physics Coordinator together with the Computing Coordinator, a persistence-capable version of Atlfast became a centerpiece of the "Lund program." This required keeping up with a rapidly-evolving target, as fast simulation developers scrambled to enhance the transient Atlfast code to support Lund physics needs, releasing seven revised versions of Atlfast in the final release preparations, in some cases involving substantial changes to the persistent data model.
With the Lund release now in bug-fixing stages, we expect to transfer responsibility for Atlfast persistence to the Atlfast package developers themselves. In the coming quarter, however, we will provide whatever support is needed to allow the Lund physics agenda to succeed.
At the beginning of the quarter the U.S.-developed Objectivity conversion service for Athena could support relatively complex objects, but inter-object references were supported only inside system-supplied event headers. Physicist-defined objects could not point to one another. While a final implementation must involve the still-developing StoreGate, in the Lund Release it was necessary to provide interim support for such pointers in the database to meet immediate needs. The Athena Objectivity conversion service was enhanced, therefore, to support Gaudi's SmartRefs. These are currently available for general use, and are used in Atlfast persistence, but are not being widely advertised since they represent an interim solution pending an implementation compatible with StoreGate.
A joint event model/database workshop was held at Argonne in early April to ensure synchronization of StoreGate and database plans. We expect the work on SmartRef persistence to transfer well to the StoreGate DataLink/DataBucket model. Database support for StoreGate DataLinks and DataBuckets is a principal deliverable expected in the next major release of ATLAS software.
Much work was done early in the quarter on the problem of bridging the gap between Athena and simulation framework (FADS/Goofy) representations of Monte Carlo events and Monte Carlo event collections. This work is currently on hold pending resolution of (non-database) issues on the convergence of Athena and FADS/Goofy. It also awaits migration of the ATLAS generator packages to a newer version of HepMC events, with CLHEP-compatible naming.
A substantial amount of U.S. database effort in the current quarter has been directed toward laying the groundwork for development of an ATLAS database architecture document. This has to date been largely an Argonne/University of Chicago activity. RD Schaffer, the European half of the ATLAS database leadership, was invited to come to the U.S. in June to work on the architecture definition, but was unable to travel. Several U.S. core database developers will instead spend the week of 16 July at CERN to accomplish this collaboration. A draft architecture document is intended to be ready for distribution by late summer.
A prototype ROOT conversion service was developed at Brookhaven, leveraging the STAR implementation of ROOT-based I/O, and will appear in an upcoming release. Work is currently underway to implement this I/O service in the ATLAS data definition language back end.
The MySQL-based implementation of liquid argon test beam conditions data developed at BNL was accepted for production use in this year’s test beam. The API was improved following input from the LAr community. The code will appear in an upcoming release. MySQL services available to ATLAS generally are also being provided by Brookhaven.
Access to raw event data information has recently been improved. The LAr group at BNL has developed converters that allow reading raw LAr hit information from Zebra, Objectivity and Root. This information is made available as a collection of hits in the TDS. Algorithms that make use of this information do not have any knowledge of the data source.
Two abstracts written and submitted by U.S. core database developers have been accepted in this quarter for the Computing in High Energy and Nuclear Physics Conference 2001 in Beijing (CHEP'01): “The ATLAS Data Management Architecture” by Malon et al and “Grid-Enabled Data Access in the ATLAS Athena Framework” by Malon et al.
Major deliverables in the next quarter are expected to include the database architecture document mentioned above, database planning for Data Challenges 0 and 1, and database support for the new StoreGate transient store in Release 3.0.0. Conversion service support for input to the detector data service, and a collection registration prototype, are also expected, as are a number of pieces of database infrastructure that are primarily the responsibility of Orsay but which will require some corresponding U.S. effort. Migration to Objectivity 6, which will require significant changes to the ATLAS build infrastructure, will also occur in the coming quarter. One of the major uncertainties in core database planning in the near term is the required support for pileup, since the scope of the core database component of this effort depends strongly on the transient computing model for handling pileup events.
WBS 22.214.171.124 Distributed Data Management and Processing Software
In a joint project with the Particle Physics Data Grid (PPDG) we initiated the development of a production distributed data service to be developed and deployed to users in a first version over the next year. The production distributed data service is to exist between CERN, the ATLAS Tier 1 facility at BNL, and a number of US ATLAS grid testbed sites (some or all of ANL, LBNL, Boston University, Indiana University, U Michigan, U Oklahoma, and UT Arlington). The objective is a multi-point U.S. Grid (in addition to the CERN link) providing distributed data services as early as possible.
Development of the distributed data manager was initiated during the period, based on the DBYA prototype introduced in the last report as a starting point. DBYA provides ATLAS-specific infrastructure and metadata, and the infrastructure to vertically integrate the data manager from the grid toolkit components employed through to the ATLAS-specific interfaces by which the system will be used. Grid toolkit components will be progressively integrated into the system. During the period a facility to load the file replica metadata of the system into the Globus replica catalog was developed. Over the next few months we expect to extend the functionality of the system from the present passive cataloging to active replication of files between grid sites using GridFTP.
In grid-related database work, a Globus replica catalog running at Argonne with catalog query and update clients running at Brookhaven has been successfully demonstrated and exercised, both from a command line interface and from C programs. This is a first step in grid-enabling data access from Athena.
Core software efforts continued to be complemented by much (primarily off-project) development activity in subsystem software. A focus of simulation activities towards the end of the period was in updating the detector geometry description in the production simulation as an early preparatory step for simulation production later this year directed at the coming Data Challenges. U.S. ATLAS members were involved in the preparation of ATLAS input to the Geant4 review which took place in June.
A facility which produces nightly builds of ATLAS software based on the most recent tagged versions of packages was developed and released. It provides immediate feedback to developers throughout ATLAS on newly introduced software bugs and incompatibilities. At present it is successfully operating at the BNL Tier I Center. At the request of the ATLAS community the facility is being installed at CERN. At the end of the reporting period the facility at CERN was under test. The facility will continue to be developed as a framework for the deployment of software test and validation tools in addition to basic build services.
The US ATLAS software librarian commenced participation in the development of a new ATLAS release management scheme based on the new CMT tool recently adopted by ATLAS.
The environment setup scripts for users of Tier I center were substantially improved. The environment is automatically tuned for particular ATLAS releases. Substantial efforts ramped up at several sites (LBNL, Indiana, and Boston U) in enabling routine remote software installation. Progress is being made through ongoing dialogue between the maintainers at these sites and the US ATLAS and ATLAS librarians. At the end of the period the development of a tool for simple, automated installation of software at distributed sites was initiated at BU.
Online web-based lectures on the use of Geant4, the production of which was described in the last monthly report, were completed and made available in June. They were recorded at a workshop held at the University of Michigan in February. The lectures were given by Andrea dell'Acqua. The course consists of 16 lectures, most between 20 and 30 minutes in length. The lectures are available on the web from the Web Lecture Archive Project server at the University of Michigan:
Slide changes are automatically synchronized to the audio and video via the Syncomat tool developed at Michigan. The intent of putting these lectures on the web is to make it possible for anyone in ATLAS who needs to use Geant4 to be able to learn to do so quickly and easily. The complete set of lectures is also available on a self-contained CD.
The XProject software project management tool developed by US ATLAS was in active use by International ATLAS for ATLAS-wide project planning by the end of the period. The new capability of the tool to ‘project’ custom WBS views from different WBS’s, developed to meet the needs of International ATLAS, was employed to create a ‘Grid WBS projection’ showing the makeup of the US ATLAS grid program in terms of its software and facility project components. This has helped illustrate and clarify the relative roles of the two projects in the grid effort, and is now being used as the basis for grid planning. Grid scheduling was also integrated into XProject.
A draft software agreement for UC Santa Cruz participation in the development of the Atlantis graphics software package was developed. We will proceed with this and other software agreements only after the pending agreement on core framework activities has been signed.
Atlfast fast simulation (Fortran and OO versions) integrated and released (Jun)
Lightweight Athena installation implemented (Jun)
Prototype Data Dictionary implementation released (Jun)
Database infrastructure supporting Atlfast persistency released (Jun)
Support for inter-object references released (Jun)
DataLinks implementing persistable relations between objects released (Jun)
ATLAS nightly build facility released (May)
Web-based Geant4 lectures completed and made available (Jun)