Although there are a lot of benchmarks with varying projection types and complexity, all are over relatively small collections. In mongo-perf:

WideProjectionTopLevelField (inclusion, exclusion, 9 projections, 10K documents)

WideProjectionNestedField (nested field inclusion, exclusion, 9 projections)

For all of the bellow: single projection with dotted path of length 3 or 6, 10K documents:

FindExclusion.ProjectionDottedField.SinglePathThreeComponentsNestedArraysOfSizeFive

FindExclusion.ProjectionDottedField.SinglePathThreeComponents

FindExclusion.ProjectionDottedField.SinglePathThreeComponentsNestedArraysOfSizeOne

FindExclusion.ProjectionDottedField.SinglePathSixComponentsNestedArraysOfSizeOne

FindExclusion.ProjectionDottedField.SinglePathSixComponentsNestedArraysOfSizeFive

FindProjectionThreeFields (3 projections, 100 docs)

ProjectInclude_CollScan_LS ( 8 projections, 10-field 100K documents)

ProjectInclude_CollScan_LL (8 projections, 23-field 100K documents)

ProjectExclude_CollScan_LL (8 projections, 23-field 100K documents)

ProjectNoExpressions_CollScan_LS( 8 projections with renaming of fields, 10-field 100K)

ProjectNoExpressions_CollScan_LL( 8 projections with renaming of fields, 23-field 100K)

We will need to extend the existing mongo-perf tests to run for three different cardinalities: 100/1000, 10K/100K, and 1M documents. The small cardinality collections are useful to emphasize the optimization time in the metrics.

• For queries over 10K/100K add cardinalities of 100 and 1M documents.
• For queries over 100 documents add cardinalities of 10K and 1M documents.