SERVER-125415Backlog
Investigate TPC-H plan quality when data does not fit in cache
For our current TPC-H Locust workloads, the cost model always assumes that all data is in cache, and we thus end up in case 1 of Mackert-Lohman to estimate the number of random IOs. We should see the effect of reducing the WT cache size on plan quality & stability. If this provides significantly different results, we should consider making an actual automated variant with this smaller cache size.