1 Job submission through UNICORE 5
UNICORE [http://www.unicore.eu/] provides a seamless interface for preparing and submitting jobs to a wide variety of computing resources. It has a three-tier design:
- A UNICORE client GUI is used for the preparation, submission, monitoring, and administration of jobs.
- The Gateway is a site’Ss point of contact for all UNICORE connections. It also checks if the user’s certificate is signed by a trusted CA. Site specific information on computing resources, including the availability of applications, is provided by a Network Job Scheduler (NJS). This server dispatches the jobs to a dedicated target machine or cluster, and handles dependencies and data transfers for complex workflows. It transfers the results of executed jobs from the target machine.
- A Target System Interface (TSI) is a server that runs on the target machine interfacing to the specific batch scheduler. TSI servers are currently running on 11 machines in DEISA interfacing various batch schedulers on different platforms: CINECA (SP5), FZJ, IDRIS and RZG employ the multi-cluster LoadLeveler (LL-MC) under AIX, ECMWF run LoadLeveler (LL) under AIX, BSC runs Moab+Slurm under Linux, HLRS employs NQS II under Super-UX, CSC, EPCC (HECToR, XT4 and X2) and LRZ uses PBS Pro under Linux and CINECA (Linux cluster) and SARA run LoadLeveler under Linux.
The DEISA Grid Infrastructure based on UNICORE, the multi-cluster shared file system (GPFS) and the multi-cluster LoadLeveler which is currently employed at five DEISA sites (connected by the dotted lines). Nearly all the machines share the GPFS file system. Normally, UNICORE users connect with their client to the gateway provided by their Home Site. For instance, a CINECA user uses the UNICORE gateway and the NJS at CINECA, whereas a LRZ user connects to the gateway at LRZ directing his job to the NJS at one of the homogeneous sites. In the heterogeneous case (i.e. where remote jobs submission between sites is not feasible on batch system level) jobs can be submitted to any DEISA site via UNICORE.
From a UNICORE client running on your workstation, you will connect to the UNICORE gateway at your Home Site. In the a case were this gateway is not available, you can connect to the UNICORE gateway of any other DEISA site. In general, every NJS server is accessible from every UNICORE gateway. But particularly in the homogeneous IBM case (every site has the same computer platform providing multi-cluster LoadLeveler), you will only need to access to the NJS of the home site; then the mechanism of remote jobs submission will allow you to run on every other IBM cluster.
Figure 1 depicts the UNICORE infrastructure with regard to the fact that each one of the five sites offers a cluster of IBM Power 5/6 machines with multi-cluster LoadLeveler.
The UNICORE client allows you to create a new job or modify existing ones. They can be composed of various tasks within a complex workflow including data transfers and dependencies. You can:
- define the workflow, in a system independent manner (UNICORE creates an internal abstract representation that allows the execution of the job on any platform),
- choose the appropriate platform for the execution of each task,
- assign resources to the task within the limits that are available on the selected target system,
- submit the job, monitor its status, and control it.