(The transformation and strategy is only available for enterprise edition, of course, see this note).

Here is a scenario, that I experienced recently, where (I thought) bitmap indexes might have been useful in “multi-user OLTP” application.
This system is used by many users and they do perform many small DMLs every day (hence qualifies the “multi-user OLTP” definition). However, this system has a set of tables, which act as base tables for a search algorithm (that is implemented using PL/SQL). These base tables are refreshed daily/weekly by a batch process. Many users concurrently perform searches (using the developed search algorithm) against these base tables (but only SELECTS and no INSERTs/UPDATEs/DELETEs). The search algorithm involves many queries that access few rows but on different combinations of columns in WHERE clause.
So, even though, the overall nature of this application is “multi-user OLTP”, this feature (search) in particular, acts like a “DSS” and can be benifited by having bitmap indexes on individual columns, instead of multiple composite B*Tree indexes, having same columns but in different order.

thank you. Your version sounds reasonable, but it doesn’t fit to “generic” claim – there’s no a word about is this multi-column indexes specific only or not. My first (and the only one) guess was something like this:

drop table t cascade constraints purge;
create table t as 
select mod(rownum, 1e4) x, lpad(rownum, 10, '0') pad from dba_objects, dba_objects 
 where rownum <= 1e6;
create index t_indx on t(x);
exec dbms_stats.gather_table_stats(user, 't', estimate_percent=>null,method_opt=>'for all columns size 1');
select /*+ gather_plan_statistics */ * from t where x in (1, 2);
@x

drop index t_indx;
create bitmap index t_indx2_01 on t(decode(x, 1, x));
create bitmap index t_indx2_02 on t(decode(x, 2, x));
exec dbms_stats.gather_table_stats(user, 't', estimate_percent=>null,method_opt=>'for all hidden columns size 1')
select /*+ gather_plan_statistics */ * from t where decode(x, 1, x) = 1 or decode(x, 2, x) = 2;
@x

-------------------------------------------------------------------------------------------------
| Id  | Operation                    | Name   | Starts | E-Rows | A-Rows |   A-Time   | Buffers |
-------------------------------------------------------------------------------------------------
|   1 |  INLIST ITERATOR             |        |      1 |        |    200 |00:00:00.01 |     207 |
|   2 |   TABLE ACCESS BY INDEX ROWID| T      |      2 |    200 |    200 |00:00:00.01 |     207 |
|*  3 |    INDEX RANGE SCAN          | T_INDX |      2 |    200 |    200 |00:00:00.01 |       7 |
-------------------------------------------------------------------------------------------------

Predicate Information (identified by operation id):
---------------------------------------------------

   3 - access(("X"=1 OR "X"=2))

-----------------------------------------------------------------------------------------------------
| Id  | Operation                    | Name       | Starts | E-Rows | A-Rows |   A-Time   | Buffers |
-----------------------------------------------------------------------------------------------------
|   1 |  TABLE ACCESS BY INDEX ROWID | T          |      1 |   1000K|    200 |00:00:00.01 |     106 |
|   2 |   BITMAP CONVERSION TO ROWIDS|            |      1 |        |    200 |00:00:00.01 |       4 |
|   3 |    BITMAP OR                 |            |      1 |        |      1 |00:00:00.01 |       4 |
|*  4 |     BITMAP INDEX SINGLE VALUE| T_INDX2_01 |      1 |        |      1 |00:00:00.01 |       2 |
|*  5 |     BITMAP INDEX SINGLE VALUE| T_INDX2_02 |      1 |        |      1 |00:00:00.01 |       2 |
-----------------------------------------------------------------------------------------------------

Predicate Information (identified by operation id):
---------------------------------------------------

   4 - access("T"."SYS_NC00003$"=1)
   5 - access("T"."SYS_NC00004$"=2)

It is very important that you included a Footenote to Philosophy8. I believe that before disussing the critical difference between B-tree indexes and Bitmap Indexes, we have first to discuss about the situations where itmap indexes are dangerous to implement.

In this context, I have one question,

Are there situations in mutli-user OLTP applications where Bitmap indexes can be safely implemented so that one can start comparing their benefit versus B-tree indexes benefit?

Mohamed Houri

At least that’s what I think he meant.

Could you please clarify what do you mean by:
>It is the combination of a subset of the available bitmap indexes that offers the same degree of precision.
I’ve read it several times – and still don’t get it.