ATLAS Chimera Reconstruction Development

Goal: provide client (U.S. ATLAS Production Team) with an integrated grid-based reconstruction framework which uses core Chimera/Pegasus, DAGMan, Condor-G components, and  the functionality provided by the GRAT production environment.
Task
 		Who
 		Description
 		Status
Client requirements Rob, Kaushik Solicit, collect production requirement and specifications.

We have no formal requirements, but we can get some guidance from the production plan.

 Production specifications have been given in email exchanges
GCL cache Rob Grid Component Library cache which holds integrated grid package solutions for specific production or user frameworks.  The model for combining grid components is discussed here (powepoint slides).

Web page of the GCL cache.

 Available with Pacman 2.099 and later.
Upgrade to VDT 1.1.7 Dan, Jerry, Jens, Alain Upgrade bundles to handle VDT 1.1.7 client libraries, particularly RLS client libraries. Complete.  RLS client bundles to go with VDT 1.1.5 Client are available. 

Problem is fixed in VDT 1.1.8
Reconstruction bundle Yuri S, Jerry This is the main pacman bundle to do the reconstruction task.

See http://grid.uchicago.edu/gcl

 

A version targeted 6.0.1 available.
Prototype production bundle (useful for testing) Jerry, Dan, Rob
 		This is GCL:atlsim-chimera. It is an atlsim-based  production bundle, used for testing chimera features.  A webpage giving instructions for this is located at: atlsim-chimera.  It includes:
  • GCL:chimera-vds
  • GCL:chimera-setup
  • GCL:chimera-setup-environment
  • VDT:VDT-Client   (from version 1.1.5)
  • IU:rls-vdt-libs
  • VDT:JDK-1.3
  • GCL:atlsim.vds.template

 

 Deprecated
ATLSIM bundle Jerry This is GCL:chimera-atlsim.

Basic Chimera submit host for ATLAS simulation, based on ATLSIM 3.2.1.

Tested on Jazz.

See http://grid.uchicago.edu/gcl

 

 Finished
RLS pacman cache Saul, Nosa Supporting Nosa at ISI providing RLS cache.

Here is the iVDGL RLS Cafe project webpage, which has links to the RLS pacman cache.

 In progress;  debugging some Globus dependencies.

4/1/03: now available, working.
RLS deployment
 		Dantong, Jason Deployment of an RLS instance used by Chimera to drive the reconstruction jobs.

RLS deployed at U of C by Ed May, available for production.

grid02.uchicago.edu

4/1/03: started.

.

GRAT integration Kaushik, Dan, Horst, Wensheng Add to the core bundle GRAT script capabilities. 
  • access production catalog
  • update Magda and AMI
 Deprecated. Approach is to pull out select scripts to interact with new production database.  See Silk scripts below.

 
Silk Scripts Kaushik Provide a limited number of scripts adopted from GRAT.

See gzipped tarfile here .

Description of scripts is here.

 

 Finished
Evaluate Silk Ed Evaluate and test individual scripts; test interactions with production database.

 

 Started
Input staging Ed, Jerry, Yuri Reserve LFNs from the ~6000 partitions to be reconstructed from Kaushiks production database. From the Chimera point of view ther are several options, and we recommend the most straightforward approach to keep the reconstuction queues filled:

* A submitter identifies a chunk files (LFN's) to reconstruct

* Magda is queried to return the PFN's

* RLS is populated with these LFN and PFN mappings

* The output is specfied to be the BNL reserved area, and told to Pegasus (i.e. specify the destination pool)

* A script is run to prepare the required chimera working directories on the execution site

* A script to generate the dags

 

  
Execution Ed, Jerry, Yuri * The submission done to keep the queue sufficiently full, perhaps doing this is 100 to 200 chunks, using a simple script.

* We can avoid scheduling decisions by simply using pdsfgrid, bu, and jazz, with separate submission directories for each.

* We do not have to get fancy here to keep all the available resources working for us.

* Most likely scenario for the next week is that Yuri Fisyak will continue submission to batch queues directly; Yuri Smirnov and Ed May will get some initial Chimera submission environments working, which can be cloned by others (using Jerry's pacmanized package) if more resources become available.

 

  
Post job   * When jobs complete they should register with the production database.

* A job which (asynchronously) runs to perform some simple validation

check

* Validated output files registered with the magda.

* A clean up script

 

  
Output management Yuri F., Kaushik Yuri Fisyak estimates that only ~0.5 TB necessary to store the total ntuple output, so an output disk has been reserved at BNL for this purpose.

 

  
MySQL backend for Chimera VDC Dan
 		This task creates a MySQL database instance to record transformation and derivations. Not started; postponed.
       
Testing of packages Yuri Fisyak, Horst Download, configure, and execute test jobs.  Provide feedback to development team. Started by Horst.
       
ATLAS base libraries Yuri Smirnov Provide pacmanized versions of required ATLAS software libraries, together with the scripts to build transformations.  At present these are based on Nordugrid rpms.

Provide 6.0.x  packages when they become available.

See http://grid.uchicago.edu/Atlas-nordugrid/

 

 5.3.0 available.

6.0.2 available.

6.0.3 expected during week of April 21.
       
Transformation deployment Yuri F, S, Ed Make sure required reconstruction transformations are deployed in the correct locations at all of the execution sites.  Done by hand on LBL, BNL
ATLAS SW grid deployment on a small number of testbed sites Yuri Smirnov Ed May Install ATLAS software software in the standard, US ATLAS testbed location (ATLAS_LOCATION) Sites:

atlas.iu.edu

 
Packaging Jerry Packaging of simulation and reconstruction bundles, loading into GCL cache.

http://grid.uchicago.edu/gcl

 

 On-gong