CK Tan wrote:
If it is convenient for you, could you run my patch against the same
hardware and data to get some numbers on select for comparison? Although
we don't address updates, copy, or inserts, we are definitely getting
at least 20% improvement in scans here without poisoning the bufpool for
large tables.
Sure, here you are:
copy-cktan-big | 00:10:08.843126
copy-cktan-big | 00:10:17.767606
copy-cktan-big | 00:09:37.797059
copy_nowal-cktan-big | 00:05:12.305025
copy_nowal-cktan-big | 00:05:10.518417
copy_nowal-cktan-big | 00:05:03.472939
select-cktan-big | 00:03:27.655982
select-cktan-big | 00:01:55.496408
select-cktan-big | 00:01:31.693856
select-cktan-big | 00:01:12.705268
select-cktan-big | 00:01:12.478247
select-cktan-big | 00:01:10.866484
vacuum_clean-cktan-big | 00:03:05.340875
vacuum_clean-cktan-big | 00:01:12.428317
vacuum_clean-cktan-big | 00:01:13.179957
vacuum_clean-cktan-big | 00:01:10.438888
vacuum_hintbits-cktan-big | 00:03:58.78208
vacuum_hintbits-cktan-big | 00:04:02.515778
vacuum_hintbits-cktan-big | 00:04:19.083402
vacuum_hintbits-cktan-big | 00:04:11.170831
copy-cktan-med | 00:02:19.413484
copy-cktan-med | 00:02:22.270497
copy-cktan-med | 00:02:22.297946
copy_nowal-cktan-med | 00:01:31.192601
copy_nowal-cktan-med | 00:01:17.736356
copy_nowal-cktan-med | 00:01:32.272778
select-cktan-med | 00:00:03.774974
select-cktan-med | 00:00:01.279276
select-cktan-med | 00:00:01.297703
select-cktan-med | 00:00:01.304129
select-cktan-med | 00:00:01.297048
select-cktan-med | 00:00:01.306073
vacuum_clean-cktan-med | 00:00:01.820733
vacuum_clean-cktan-med | 00:00:01.755684
vacuum_clean-cktan-med | 00:00:01.755659
vacuum_clean-cktan-med | 00:00:01.750972
vacuum_hintbits-cktan-med | 00:00:01.58744
vacuum_hintbits-cktan-med | 00:00:06.216038
vacuum_hintbits-cktan-med | 00:00:02.201789
vacuum_hintbits-cktan-med | 00:00:36.494427
Select performance looks the same as with Simon's/my patch.
That 20% gain must be hardware-dependent. My interpretation is that the
CPU savings from more efficient cache usage only matters when the I/O
bandwidth is high enough that CPU becomes the bottleneck.
--
Heikki Linnakangas
EnterpriseDB http://www.enterprisedb.com
---------------------------(end of broadcast)---------------------------
TIP 6: explain analyze is your friend
