Details
-
Spike
-
Not Assigned
-
None
Description
As a result of our work on ExecutionBroker we have identified a critical part of offering time slots for execution tasks will involve predicting how long it will take to stage the required input data at the target platform, and staging the output results back into permanent storage. Our experience suggests that this functionality could be best implemented as a separate component, 'used by' rather than 'contained by' ExecutionBroker.
The primary function of the interface would be to predict the cost of moving data to and from a compute platform, without having to move the data.
In order to make an offer for executing a given task with a given set of input data, Executionbroker needs to find out how long it will take to stage the data from its current location to the target compute platform in preparation for executing the task.
In simple terms, "how long would it take to move <this> data to <this> site".