Re: [PERFORM] Odd performance results - more info

[Stephan Szabo]

On Tue, 5 Aug 2003, Medora Schauer wrote:

 I hope this piques someones curiosity.  I'd really like to know
 what is going on here...

I think you're getting caught by the typing of constants preventing
index scans.

  UPDATE shot_record SET trace_count = %d  \
  WHERE shot_line_num = %d  \
AND shotpoint = %d  \
AND index = %d ,
  0, shotline, shotpoint + i, 0);

I believe that the int constants are going to generally be treated as
int4.  If you're comparing them to an int8 you're not going to get
an index scan probably.  Try explicitly casting the constants to
the appropriate type: CAST(%d AS int8).


 snprintf(buffer, sizeof(buffer),
  UPDATE shot_record SET trace_count = %d  \
  WHERE shot_line_num = %f  \
AND shotpoint = %f  \
AND index = %d ,
  0, (float)shotline, (float)shotpoint + (float)i, 0);

Same general issue here, I think the floats are going to get treated
as float8 in 7.1, so you'll probably need an explicit cast.

As Joe said, try explain on the queries for more details.



---(end of broadcast)---
TIP 6: Have you searched our list archives?

   http://archives.postgresql.org

Re: [PERFORM] Odd performance results - more info

[Joe Conway]

Medora Schauer wrote:
I would greatly appreciate it if someone could run this code in their
environment and let me know if you get results similiar to mine.
The INT test results in execution times of 11 - 50+ secs increasing
each time the test is run.  The FLOAT test execution times are
consistently  3 secs regardless of how many times it is run.
Without actually trying the code, I'd bet that an index is getting used 
for the float8 case, but not in the int8 case:

if (dtype == INT_TYPE){
snprintf(buffer, sizeof(buffer),
 UPDATE shot_record SET trace_count = %d  \
 WHERE shot_line_num = %d  \
   AND shotpoint = %d  \
   AND index = %d ,
 0, shotline, shotpoint + i, 0);
} else {
snprintf(buffer, sizeof(buffer),
 UPDATE shot_record SET trace_count = %d  \
 WHERE shot_line_num = %f  \
   AND shotpoint = %f  \
   AND index = %d ,
 0, (float)shotline, (float)shotpoint + (float)i, 0);
}
Try running EXPLAIN ANALYZE on these update statements manually. It also 
might help to run VACUUM ANALYZE after populating the tables.

HTH,

Joe

---(end of broadcast)---
TIP 9: the planner will ignore your desire to choose an index scan if your
 joining column's datatypes do not match