Parallel Computing Toolbox
Schedulers manage, monitor, and distribute workload and administer resources across distributed computing systems comprising disparate hardware and software resources.
The scheduler interface provided by MathWorks parallel computing products is a high-level abstraction that lets you submit jobs to your computation resources without having to deal with differences in operating systems, environments, and schedulers.
MATLAB Distributed Computing Server provides a basic scheduler, the MathWorks job manager. It also includes direct support for Platform LSF, Microsoft Windows Compute Cluster Server 2003 (CCS), Microsoft Windows HPC Server 2008 (HPCS), PBS Pro, and TORQUE schedulers, and indirect support for others through a generic scheduler interface.
|Scheduler||Level of Support (more...)|
|MathWorks Job Manager||Included with MATLAB Distributed Computing Server|
|Platform LSF||Fully integrated|
DataSynapse also provides a fee-based consulting service.
|Grid Engine Family:
Sun N1 Grid Engine
Sample integration scripts are available with the
|Grid MP||Vendor-provided integration scripts are available online.|
User must create integration scripts using the generic API available with the products.
Licenses are checked out when MATLAB Distributed Computing Server workers are launched. With MathWorks job manager, workers are run as a service, and licenses remain checked out until workers are shut down. With other schedulers, workers are run as an application, and licenses are checked out each time a MATLAB job is launched and checked back in when the job finishes. See License Management with Third-Party Schedulers for more information.
You can switch between schedulers with minimal code changes, as shown in the code sample. You can also use configurations to customize and maintain settings such as file and path dependencies for multiple schedulers or even for multiple projects using the same scheduler.
Setting up and running an application with the MathWorks job manager or other third-party schedulers. The difference in code required for various schedulers is minimal. Click on image to see enlarged view.