Conclusions and Future Work
This framework for submitting control and managing the jobs over the glite grid infrastructure could sensibly reduce the man power needed to run large challenges. With the new features it allows the final user to deal easily with large input files anda huge collection of output file in a unattended way. It was mainly tested with bioinformatics applications but it surely could work with other applications coming from different science, thanks to it flexibility and to the possibility to be customized to meet the end users needs both for serial and for workflow driven applications.
In this work we will present the status of development of the Job Submission Tool both in terms of new functionalities and new applications ported within this framework.
The main new feature added are: automatic distribution of input data using standard gLite Data Management Tools, the possibility to exploit the dependency between tasks belonging to the same run, the possibility to create on the fly the needed task from both existing portals or application running over the grid (without any human intervention).
The Job Submission Tool is an easy and highly customizable framework that speeds up the process of porting new application to the grid glite infrastructure.
The applications submitted using this framework could be executed both using robot certificate or user certificates.
In the work we will present also the portal built on top of the framework in order to allow final users to submit large collection of jobs by filling up a small number of web form.
We will also describe the auto-adaptive submission algorithm that could guarantee to distribute the load of the calculation over all the farm available within the grid infrastructure.
Using this tool the end user could easy create a collection made up a large number of jobs, where the complete run could take months to be executed also in a large grid infrastructure like EGEE .
In this case one needs to solve problems like: detecting application failures, resubmit the failed jobs, collect the output back. Moreover in some bioinformatics applications the input files are fairly large, and this could easily become the main bottleneck if those files are available on one or a few sites. In this case it could be difficult to know where the files should be replicated as it may depend on where the CPU are available. In those cases it could be interesting to have a procedure to move the data where the CPU are free at a given time, and for the framework do this automatically without human intervention, in order to run on several farms, using data locally available.
Moreover as the number of the jobs needed to finish the whole run could easily be several thousands it is useful to be sure that the submission procedure does not overload a single or few sites and that the jobs are spread among the largest number of available sites.
|URL for further information||http://webcms.ba.infn.it/cms-software/index.html/index.php/Main/JobSubmissionTool|
|Keywords||Job Submission, Workflow, gLite, bioinformatics, Data Management|