Hierarchical Report

1. Tests at UWM suggest that template generation can cost as much as 20% that of filtering.  This strongly suggests that we need more data when mpiAPI starts a job.    As a concrete example,   we imagine for LIGO I that the data should be filtered in segments of approximately 15 minutes.   We suggest that templateFilters() should be called with 5 x (15 min segments) at a time.   This would reduce the burden of template generation to (20/5)%.
2. Within the current model for wrapperAPI,  the cost of filtering in a hierarchical search may be inreased by of order 14%.   If the baseline for a brute force search is unity (1) and the computational cost for the equivalent hierarchical search is 0.7,   then the real cost should be about 0.8.   Some of this might be recovered if the templateFilters() function is allowed to communicate with other nodes in the comm-world.
3. In the hierarchical search,  the an efficient use of templateFilters() would filter one-coarse template and if it gets a hit,   then drop onto the fine templates of the hierarchical search.    Thus different instances of templateFilters() might take significantly different times to complete.   This should be accounted for in the load balancing stage.