When implementing the CVE data model, performance optimizations from the old model were skipped hoping they were unnecessary.  After running a long running cluster for most of a week, it appears that we should continue using those optimizations.  In this instance, the optimization that only does the necessary aggregation for sorting instead of all the aggregations and then perform the subsequent aggregations on a second pass.

The performance of the new without this optimization is slightly worse than the performance of the old with this optimization.