histogram for read/write times
This feature was suggested by Jim Browne. It would be helpful to have a histogram not only of the access sizes, but how long each access took to complete.
We already time each read and write (in order to record cumulative I/O time, and to identify the slowest operation), so there wouldn't be any extra overhead except for the storage space required for the new fields.
We would want the bucket sizes to scale automatically based on the slowest operation we have seen on that file.