MohammedB wrote:
> I'm having a little trouble trying to conceptulize this scenario.
>
> I've read that index use is optimal when selecting 5% or less records
> in a table. I would have thought that anything less than 49% would be
> effective since you'd have one read of the index and one read to
> retrieve the corresponding record (reads < 98% of total data). I'm
> talking about a basic equality query using b-tree indexes and no PQ.
> Would'nt this be more efficient than a full-table scan (reads = 100% of
> data)? Also, why is the break even point around 5%? Is their some
> sort of optimization theory involved here?
>
- you do not take into account b*tree traversing, but, more important
- indexed lookups are done through monoblock I/Os, whereas FTS is done
through multiblock I/O (db_file_multiblock_read_count blocks for each
FTS I/O). The estimated cost is chiefly derived from the overall number
of I/Os.
Received on Tue Jan 18 2000 - 04:49:12 CST
