Reads on "Mined" are a bit slower on average than "MinedPast" in perm storage

[carneiro-cw]

Unexpected, Mined should be considerably faster.

[carneiro-cw]

Maybe if we lower level0_file_num_compaction_trigger we can get better results. All files in l0 must be read since they are not ordered yet (although we do have bloom filters to theoretically greatly reduce l0 reads)