Yes, doing a copy with PAGE_SIZE of 2k reduced the average query time even
a bit more, to under 100ms. So this is mostly about the order of the data
on disk.
On Wednesday, October 9, 2013 10:48:14 PM UTC-7, Noel Grandin wrote:
>
>
> It's quite likely that a dump and restore improved the locali
It's quite likely that a dump and restore improved the locality of
reference because all of the data belonging to a single table is now
packed tightly together.
On 2013-10-10 06:04, Brian Craft wrote:
Copying the db to a new db with PAGE_SIZE of 8k lowered the average
query time by a facto
Continued posting in my own thread, in case it's useful to anyone. ;)
Copying the db to a new db with PAGE_SIZE of 8k lowered the average query
time by a factor of 20: from 10 sec to 0.5 sec.
I should copy it to a db with the original PAGE_SIZE to see if insert order
affects anything (since the
After more investigation I strongly suspect that IO is my bottleneck, but
the disk read rate is low because there are very many seeks being generated
by h2.
I expect PAGE_SIZE is the best way to address this. Is that correct?
Does row insert order affect this much? For example, would alternate