Next: Conclusion
Up: Parallelizing Classification
Previous: Finding Split Points
- Since the processor can have attribute records belonging to any leaf,
the rids need to be collected from all processors for building the
probe structure.
- After partitioning the list of a leaf's splitting attribute, the rids
collected during the scan are exchanged with all other processors.
- After the exchange, each processor continues independently,
constructing a probe-structure with all the rids and using it to
split the leaf's remaining attribute lists.
- In all other respects the parallel splitting is same as in the serial
case.
DBMS
1999-03-11