CERN Accelerating science

This website is no longer maintained. Its content may be obsolete. Please visit http://home.cern/ for current CERN information.

Summary of meeting with LHCb on 13/12/02

Present:
Guy Barrand, Nick Brook, Marco Cattaneo, Philippe Charpentier, Gloria Corti, John Harvey, Pere Mato

Distributed Analysis

Current approach

To manage wide-area distributed computing LHCb has develop SLICE.
SLICE addresses large batch production: it is based on WAN and LAN client-server architectures for job submission and control. It integrates a work-flow manager based on RDBMS technology.
In many respects SLICE is similar to Alien even if it is based on very different technologies.

Short-Medium term plans

At the moment Slice does not address end-user needs besides publishing a "catalog" of the available data.
There is no plan to expand SLICE to cover non-production use-cases. For this LHCb is developing, together with ATLAS, Ganga.

LHCb does not plan any tight integration neither in Slice nor in Ganga of grid-middleware (at least in the time scale of LCG-1). LCG-1 will be seen just as one more site available to LHCb.

LHCb is worried about the robustness, readiness and stability of grid middleware, particularly for what concern the content of LCG-1.

Interactive Environment

Current approach

LHCb simulation, reconstruction and analysis software is based on the Gaudi framework. The current approach sees simulation and reconstruction run as production jobs with output a "DST" containing reconstructed objects.
Physicists run an analysis program (DaVinci) to produce analysis objects or directly histograms.
Analysis objects are saved using standard Gaudi persistency-service. Currently, transient histograms are based on AIDA/HTL and saved as native Root histograms by the Gaudi histogram-service.

Each event is annotated with a tag that is used for fast selection. Physicists currently use an implementation based on Root trees. A RDBMS based implementation (using an ODBC interface) is also available.

Medium/Long Term view

LHCb general strategy is to provide a palette of possibilities centred around python as component glue.

Although root-trees may well remain the most popular implementation of event-tags, LHCb does not plan to use native root as analysis environment: Event-tags will be used solely for fast selection purposes, to store any long-life quantity physicists will be encouraged to create analysis-objects.

LHCb plans to deploy an interactive version of DaVinci with the ability to create and access analysis-objects.
The baseline implementation sees a command-line interface based on python. Access to analysis-objects and histogram filling is supposed to be performed in standard C++ Gaudi modules eventually edited, compiled and dynamically loaded interactively. Analysis-object may also directly accessed from python using the lcg-dictionary.
In this environment Root (as any other analysis tool such as Exel or Mathematica) will be accessed through a python gateway and used mainly for visualisation purposes.

An alternative approach may use Root as user interface. Gaudi modules will be dynamically loaded and full access to analysis-object provided either through C++ compiled modules or directly in CINT using a gateway between the LCG-dictionary and CINT.

AIDA

Gaudi currently uses AIDA-2 as interface to histogramming services. No experience exists w.r.t. AIDA-3.
AIDA-2 is seen mainly as an abstract interface among frameworks (Gaudi and Root in the specific case).

LHCb considers of fundamental importance the development not only of a common interface to data analysis but also of proper implementations independent of any particular analysis framework.

Python

Python is currently considered the baseline scripting language in LHCb to be used in all environments: from the development of scripts to manage operating system services to its use as command-line physicist interface.

Event Display

LHCb event display, Panoramix, offers both full 3D and specialised 2D graphics based on OpenInventor. The GUI, Based on the OnX framework, is described using XML and uses Python to implement actions.

Final Consideration

LHCb plans to keep (at least in the short-term) a clear separation between production and analysis environments both for what concern tools and deliverables persistent objects.
It should be noted that all applications are based on the same framework, Gaudi, and use the same services. Therefore migration of code between the two environments will be easy.

LHCb considers critical to provide easy access (as easy as sending an e-mail) to grid resources: Ganga has been designed with this very goal in mind.

LHCb is convinced that Python is an optimal environment to provide both a standard and consistent user interface and an integration-glue among independent components.

LHCb considers to be in the scope of LCG the task of providing analysis-tools that are independent of any particular analysis framework.

Possible common projects:

References

Computing includes direct links to Gaudi, Ganga, DaVinci and Event-Display
Data Challenges includes links to Slice and to plans for 2003

Vincenzo Innocente
Most recently modified on Mon Jan 13 09:44:56 MET 2003 by Vincenzo Innocente