#2 readahead small maps first

[Anonymous]

Hello,

I think the first slowdown in the process of reading a file from the disk is triggering the disk move up to the physical position of the file.

The file size is a secondary slowdown.

If you consider that you need to read the content of a file in order to know what is the next file to upload (library dependency), then I think the number of files loaded into memory is more important than the total size of the files loaded into the memory.

Therefore, I think you could speed-up preload if you choose to preload small files first in addition to the statistical forecast.