Suggests for a more efficient query?

Question

I have a client (customer) that uses a 3rd party software to display graphs of their systems. The clients are constantly asking me (the DBA consultant) to fix the database so it runs faster. I've done as much tuning as I can on the database side. It's now time to address the application issues. The good news is my client is the 4th largest customer of this 3rd party software and the software company has listened and responded in the past to suggestions.
All of the tables are setup the same with the first column being a DATE datatype and the remaining columns are values for different data points (data_col1, data_col2, etc.). Oh, that first date column is always named "timestamp" in LOWER case so got to use double quotes around that column name all of the time. Each table collects one record per minute per day per year. There are 4 database systems, about 150 tables per system, averaging 20 data columns per table. I did partition each table by month and added a local index on the "timestamp" column. That brought the full table scans down to full partition index scans.
All of the SELECT queries look like the following with changes in the column name, table name and date ranges. (Yes, we will be addressing the issue of incorporating bind variables for the dates with the software provider.)
Can anyone suggest a more efficient query? I've been trying some analytic function queries but haven't come up with the correct results yet.
SELECT "timestamp" AS "timestamp", "DATA_COL1" AS "DATA_COL1"
FROM "T_TABLE"
WHERE "timestamp" >=
(SELECT MIN("tb"."timestamp") AS "timestamp"
FROM (SELECT MAX("timestamp") AS "timestamp"
FROM "T_TABLE"
WHERE "timestamp" <
TO_DATE('2006-01-21 00:12:39', 'YYYY-MM-DD HH24:MI:SS')
UNION
SELECT MIN("timestamp")
FROM "T_TABLE"
WHERE "timestamp" >=
TO_DATE('2006-01-21 00:12:39', 'YYYY-MM-DD HH24:MI:SS')) "tb"
WHERE NOT "timestamp" IS NULL)
AND "timestamp" <=
(SELECT MAX("tb"."timestamp") AS "timestamp"
FROM (SELECT MIN("timestamp") AS "timestamp"
FROM "T_TABLE"
WHERE "timestamp" >
TO_DATE('2006-01-21 12:12:39', 'YYYY-MM-DD HH24:MI:SS')
UNION
SELECT MAX("timestamp")
FROM "T_TABLE"
WHERE "timestamp" <=
TO_DATE('2006-01-21 12:12:39', 'YYYY-MM-DD HH24:MI:SS')) "tb"
WHERE NOT "timestamp" IS NULL)
ORDER BY "timestamp"
Here are the queries for a sample table to test with:
CREATE TABLE T_TABLE
( "timestamp" DATE,
DATA_COL1 NUMBER
INSERT INTO T_TABLE
(SELECT TO_DATE('01/20/2006', 'MM/DD/YYYY') + (LEVEL-1) * 1/1440,
LEVEL * 0.1
FROM dual CONNECT BY 1=1
AND LEVEL <= (TO_DATE('01/25/2006','MM/DD/YYYY') - TO_DATE('01/20/2006', 'MM/DD/YYYY'))*1440)
Thanks.

Answer 1

No need for analytic functions here (they’ll likely be slower).
1. No need for UNION ... use UNION ALL.
2. No need for <quote>WHERE NOT "timestamp" IS NULL</quote> … the MIN and MAX will take care of nulls.
3. Ask if they really need the data sorted … the s/w with the graphs may do its own sorting
… in which case take the ORDER BY out too.
4. Make sure to have indexes on "timestamp".
What you want to see for those innermost MAX/MIN subqueries are executions like:
03:19:12 session_148> SELECT MAX(ts) AS ts
03:19:14   2  FROM "T_TABLE"
03:19:14   3  WHERE ts < TO_DATE('2006-01-21 00:12:39', 'YYYY-MM-DD HH24:MI:SS');
TS
21-jan-2006 00:12:00
Execution Plan
   0   SELECT STATEMENT Optimizer=ALL_ROWS (Cost=2.0013301108 Card=1 Bytes=9)
   1    0   SORT (AGGREGATE)
   2    1     FIRST ROW (Cost=2.0013301108 Card=1453 Bytes=13077)
   3    2       INDEX (RANGE SCAN (MIN/MAX))OF 'T_IDX' (INDEX) (Cost=2.0013301108 Card=1453 Bytes=13077)