Environment from the Molecular Level
A NERC eScience testbed project
My_condor_submit (MCS) is our primary grid job submission tool. The tool itself is a perl program that uses the Condor-G wrapping of Globus together with a simple workflow (based on Condor's DAGman tool) to incorporate data management.
MCS requires the user to prepare a relatively simple job description document loosely based on standard Condor scripts. It was found early in the eMinerals project that scientist users can generally work quite comfortably with Condor scripts. Simple and more complex example MCS scripts can be found from these links (example 1 and example 2).
Basic operation model
MCS is designed to work hand-in-hand with the Storage Resource Broker (SRB). In its original form, MCS operated as a three stage process. Before submitting a job, the user would place all his/her input files into a specified directory within the SRB. MCS would then follow the sequence:
- On execution, MCS initially downloads all relevant files (input data files and executable if desired) from the SRB to the Globus gatekeeper for the' grid computing resource being used (by submitting a small perl script call pre.pl as a Globus fork job).
- MCS will then run the computational job through the remote job management system using the downloaded files as input files for the main calculation.
- At the end of the job, MCS will run another small perl script (called post.pl) which will upload generated files from the remote gatekeeper to one of the SRB vaults.
After a job has completed, the user has a complete archive of all input and output files within the SRB.
MCS has the following features
- Close interaction with the SRB
- Ability for users to specify multiple SRB directories (called collections) for files
- MCS can submit jobs to Globus gatekeepers that manage access to a wide range of grid resorces, including Condor pools and clusters with a variety of job managers (including PBS, SGE and IBM LoadLeveller).
- Generalised to support submission of jobs to resources that are not part of the eMinerals minigrid.
- Metascheduling with load balancing across all minigrid and external resources.
- Automatic collection of metadata from output XML files, using the AgentX and RCommand tools.
- MCS enables users to add arbitrary metadata to output files.
MCS must be launched from a computer on which Condor and Globus are installed. Users will need to have X.509 digital certificates which will be recognised by the compute resources being accesses. Users need to have SRB accounts, and in order to use metadata they also need to be registered with the RCommand databaser server.
Clearly it is also necessary for the remote resources to have SRB client tools, the RCommand client tools and the AgentX libraries installed in order to access all facilities.
The SRB and RCommand client tools require that the user has created some configuration files. We have tools to enable users to create these files remotely.
MCS was originally designed to be used one job at a time. However, there are two important adaptations.
- We have incorporated MCS into a combinatorial job production and management package, in which one job description document is created for each job within a parameter sweep.
- We have developed a web-services wrapping around MCS to enable MCS to be launched from a computer on which neither Globus nor Condor are installed.
Papers that describe the MCS and job submission issues within the eMinerals project can be obtained as pdf downloads from the following links:
Collaborative grid infrastructure for molecular simulations: The eMinerals minigrid as a prototype integrated compute and data grid.
M Calleja, R Bruin, MG Tucker, MT Dove, RP Tyer, LJ Blanshard, K Kleese van Dam, RJ Allan, C Chapman, W Emmerich, PB Wilson, JP Brodholt, A Thandavan, VN Alexandrov.
Molecular Simulations 31, 303313, 2005
eMinerals: Science Outcomes enabled by new Grid Tools.
M Alfredsson, E Artacho, M Blanchard, JP Brodholt, CRA Catlow, DJ Cooke, MT Dove, Z Du, NH de Leeuw, A Marmier, SC Parker, GD Price, JMA Pruneda, W Smith, I Todorov, K Trachenko, and K Wright.
Proceedings of All Hands 2005 (ISBN 1-904425-53-4), pp 788795, 2005
Last edit 11/7/06
Copyright and contact information here