Ontology alignment consists of many mapping tasks, for example, the discovery of parent-child concept pairs, the finding of sibling concept pairs, etc. OAANN concentrates on finding pairs of equivalent concepts as the first step.
Nevertheless, for NoC-based reconfigurable systems, the issue of mapping tasks to the PEs becomes more complex, due to the characteristic of hardware reconfiguration.
The map tasks of iteration i+1 start their execution and process their data, even while some of the reduce tasks of iteration i are still processing their input.
When any reduce task of iteration i completes and produces its output, it sends the output immediately to the map tasks of iteration i+1.
In this case, the map tasks of the next iteration cannot start their execution until all reduce tasks of the previous iteration are completed.
Later, the updated dynamic data from the reduce tasks are passed to the nodes that hold their static data to start the map tasks of the next iteration.
We map tasks of applications onto an 8 × 8 NoC-based MPSoC to show that our mapping heuristics consistently leads to reduction in the total execution time, energy consumption, average channel load and latency.
The reduce tasks take the updated intermediate results from the merge operation as input, recompute and update the dynamic data, and then pass them to the map tasks of the next iteration.
iiHadoop further improves the performance by efficiently overlapping the time of the reduce and map tasks of two consecutive iterations, and using Hash Join algorithm to join the static and dynamic data as discussed in "Iterative processing" section.
When (Map_{0}) finishes processing its input, (Reduce_{0}) and (Reduce_{1}) start collecting intermediate data from (Map_{0}), and start the execution of the reduce function when all map tasks of iteration i+1 complete their execution (e.g., (Map_{1})).
At each iteration, our MapReduce implementation generates different number of map tasks for different input size.
