"Mark Liberman" <mliberman(at)mixedsignals(dot)com> wrote
>
> Now, my follow-up question / assumption. I am assuming that the IO time
> is
> so long on that index because it has to read the entire index (for that
> file_id) into memory
>
> any confirmation / corrections to my assumptions are greatly appreciated.
> E.g. is
> the compound index the way to go, or the solo index on file_id?
>
Only part of the index file is read. It is a btree index. Keep the index
smaller but sufficient to guide your search is always good because even by
the guidiance of the index, a heap visit to get the real data is not
avoidable.
Regards,
Qingqing