Re: [PERFORM] query performance with hstore vs. non-hstore

2014-09-01 Thread David G Johnston
Huang, Suya wrote > See output of explain (analyze,timing off), the total runtime is close to > the one enable timing. Calling 43s "close to" 70s doesn't sound right... > dev=# explain (analyze, timing off) select cha_type, sum(visits) from > (select (each(visits)).key as cha_type,(each(visits))

Re: [PERFORM] query performance with hstore vs. non-hstore

2014-09-01 Thread Huang, Suya
Hi Pavel, See output of explain (analyze,timing off), the total runtime is close to the one enable timing. dev=# EXPLAIN (ANALYZE, TIMING OFF) select cha_type,sum(visits) from weekly_non_hstore a join seg1 b on a.ref_id=b.ref_id group by cha_type order by sum(visits) desc;

Re: [PERFORM] query performance with hstore vs. non-hstore

2014-09-01 Thread Pavel Stehule
2014-09-01 8:54 GMT+02:00 Huang, Suya : > Thank you Pavel. > > > > The cost of unpacking hstore comparing to non-hstore could be calculated > by: > > Seq scan on hstore table + hash join with seg1 table: > > Hstore: 416.741+ 34619.879 =~34 seconds > > Non-hstore: 8858.594 +26477.652 =~ 34 seconds