??? wrote: > For example, tableA contains two columns: implicit rowid, A_id. > we create index on A_id. > firstly, we used rowid to select the row, cost about 400 seconds; > secondly, we used A_id to select the row, cost about 200 seconds; > thirdly, we also used rowid to select the row, cost about 200 seconds.
This can be explained by the data being cached. > if firstly,we used A_id to select the row, cost about 200 seconds; > the create index on A_id may have some influence to do with the rowid. > and from this example, I also think the rowid is not implicit index. If all columns of the table are contained in the index, then the "select *" can read all values from the index itself, without having to look up the row in the original table. This is called a covering index. If your table has more columns, lookups on A_id will become slower. Regards, Clemens
