The page presents current status of the remote/distributed/parallel plug-in execution.
1. We have a separate branch for our tests at https://svn.gforge.hlrs.de/svn/larkc/branches/distributed
2. We agreed on the implementation strategy for the plug-in manager
- IBIS-containers based on JavaGAT
3. We are investigating possibilities of the remote plug-in containers execution
we created a plug-in manager supporting the distributed plug-in execution, eu.larkc.core.pluginManager.distributed
we created an IBIS containers for the distributed execution of the Reasoner plug-in eu.larkc.core.pluginManager.distributed.?ReasonerManager. The methods for uploading code and data (prepareSubmission()), remote container execution and produced data dowloading (runRemoteJob()) are already implemented, but not tested yet (some platform fixes are required)
we set up a test pipeline where the reasoner can be executed remotely (eu.larkc.core.Platform)
- we're trying to execute the pipeline in a distributed environment including high-performance resources (the platform is running on the user's local machine, the pipeline is started and executed on a user's local machine, while the reasoner runs on the HLRS cluster). But some local tests are still required...
4. Main further actions:
- we will develop distributed containers for other plug-in types (Identifier - partially done, Decider, Selector, Transformer)
- we will elaborate a strategy how to deal in case of dynamic data flow (implementing containers for a real pipeline)
- we will implement a real pipeline for distributed execution - the main outcome of our activities for the 2nd review
- we will adapt the developed mechanisms for different types of distributed environments (high-performance grid, desktop grid, etc.)
- we will parallelize several plug-ins (and probably pipelines...) in order to get maximal performance of the execution(the first candidate - the DIG reasoner of the VUA).