RE: Better SQL performance with UNION?

Vern, thanks for taking a kick at the can; I'm not surprised at your
results.  I tried CREATE INDEX on each of the columns and (with STRDBG) and
the Optimizer indicated it used the index.  It's time for a real test, I
guess...

I thought about EVI's but I thought I read (in the "Query Performance and
Optimization" manual?) they're better on columns with a relatively small
number of values.

In thinking about an alternative solution, I've come up using with a trigger
on the PF to write cross-reference entries to a finder file.  But it's so
"System/36"-like it makes me gag.  It could be ITER...

Thanks,
reeve

-----Original Message-----
From: midrange-l-admin@midrange.com [mailto:midrange-l-admin@midrange.com]On
Behalf Of Vernon Hamberg
Sent: Saturday, September 21, 2002 2:17 AM
To: midrange-l@midrange.com
Subject: RE: Better SQL performance with UNION?

Well, I tried some of my own suggestions. I had three columns I wanted to
include, along with a customer key field. I had the longest field in the
first SELECT of the UNION. No matter what I used, it chose arrival
sequence. When I ran it with no indexes available, the optimizer suggested
building indexes on the respective 2nd column. I tried this, both with
normal indexes and with EVIs. The estimated time with no indexes was 20
seconds. With normal indexes, 21 seconds. With EVI, 29 seconds.

  I also tried indexes on the respective column 2 and the customer key,
hopin it'd do an index-only access method. Nope. Same time estimates.

It was interesting that, with EVIs, the first SELECT was executed in
arrival sequence with bitmap processing, meaning it used the EVI. But the
2nd and 3rd SELECTs did not use bitmap processing. Curious.

Results were the same, whether I used a compare value early in the alphabet
or one that was quite high in the alphabet.

Using DEGREE(*OPTIMIZE) was faster - 2 of the selects used bitmap
processing with EVIs. Estimate was 15 seconds.

Better yet was when the keys were normal indexes, of, e.g., C_NAME,
C_CUSTKEY. The first select was processed using the logical, index-only.
The second used the logical, not index-only, the third one was arrival
sequence. Estimated time, 12 seconds, for a search starting with 'X', 14
seconds for one from 'D'.

So the best results should come from using indexes that support each SELECT
in the UNION. These should be normal (not EVI) indexes, each with 2 fields,
the first being the WHERE clause field and the second being cmcust. Query
attribute DEGREE should be *OPTIMIZE or *MAX.

BTW, the OPTIMIZE FOR n ROWS clause had no effect.

Test file has about 150,000 records, record length is 223, columns used
were C_NAME (25 char), C_ADDRESS (40 char), and C_PHONE (15 char), plus the
common field C_CUSTKEY (BIN 4)

Bottom line - it was slow. Machine is a 270-2248 with 1517 interactive
feature

Regards

Vern

_______________________________________________
This is the Midrange Systems Technical Discussion (MIDRANGE-L) mailing list
To post a message email: MIDRANGE-L@midrange.com
To subscribe, unsubscribe, or change list options,
visit: http://lists.midrange.com/cgi-bin/listinfo/midrange-l
or email: MIDRANGE-L-request@midrange.com
Before posting, please take a moment to review the archives
at http://archive.midrange.com/midrange-l.