Hi Soumyadeep,
Thanks for re-running the tests.
On Thu, 23 Jul 2020 at 06:01, Soumyadeep Chakraborty
<soumyadeep2007@gmail.com> wrote:
> On Tue, Jul 14, 2020 at 8:52 PM David Rowley <dgrowleyml@gmail.com> wrote:
> > It would be good to see EXPLAIN (ANALYZE, BUFFERS) with SET
> > track_io_timing = on; for each value of max_parallel_workers.
>
> As for running EXPLAIN ANALYZE, running that on this system incurs a
> non-trivial amount of overhead. The overhead is simply staggering.
I didn't really intend for that to be used to get an accurate overall
timing for the query. It was more to get an indication of the reads
are actually hitting the disk or not.
I mentioned to Kirk in [1] that his read speed might be a bit higher
than what his disk can actually cope with. I'm not too sure on the
HDD he mentions, but if it's a single HDD then reading at an average
speed of 3457 MB/sec seems quite a bit too fast. It seems more likely,
in his cases, that those reads are mostly coming from the kernel's
cache.
David
[1] https://www.postgresql.org/message-id/CAApHDvoDzAzXEp+Ay2CfT3U=ZcD5NLD7K9_Y936bnHjzs5jkHw@mail.gmail.com