On Wed, 2009-05-27 at 21:09 -0400, Tom Lane wrote:
So, if we assume that these numbers are real and not artifacts, it seems
we have to postulate at least four distinct block-size-dependent
performance effects:
Two performance effects would be sufficient to explain the results.
* Optimal
On Wed, 2009-05-27 at 17:51 -0700, Mark Wong wrote:
On Wed, May 27, 2009 at 1:46 AM, Simon Riggs si...@2ndquadrant.com wrote:
On Tue, 2009-05-26 at 19:51 -0700, Mark Wong wrote:
It appears for this workload using a 16KB or 32KB gets more than 4%
throughput improvement, but some of that
On Tue, 2009-05-26 at 19:51 -0700, Mark Wong wrote:
It appears for this workload using a 16KB or 32KB gets more than 4%
throughput improvement, but some of that could be noise.
The baseline appears to have a significant jump in txn response time
after 77 mins on the baseline test. I think you
Why on earth would there be a dip precisely at 8k with both smaller
and larger block sizes being faster??
--
Greg
On 27 May 2009, at 03:51, Mark Wong mark...@gmail.com wrote:
Hi all,
A long long time ago (in 2004) I ran a series of tests surveying the
results of changing BLCKSZ when it
On Wednesday 27 May 2009 05:51:05 Mark Wong wrote:
BS notpm % Change from default
-- - --
1 14673 -4.8%
2 15864 2.9%
4 15774 2.3%
8 15413 (default)
16 16118 4.6%
32 16051 4.1%
64 14874 -3.5%
This means that both somewhat larger and somewhat smaller than 8 give better
On Wed, May 27, 2009 at 1:46 AM, Simon Riggs si...@2ndquadrant.com wrote:
On Tue, 2009-05-26 at 19:51 -0700, Mark Wong wrote:
It appears for this workload using a 16KB or 32KB gets more than 4%
throughput improvement, but some of that could be noise.
The baseline appears to have a
Mark Wong mark...@gmail.com writes:
Oopsies. I've rerun, but now that there is no dip, the average
throughput still didn't change much:
BS notpm % Change from default
-- - --
1 14673 -5.1%
2 15864 2.7%
4 15774 2.1%
8 15454 (default)
16 16118 4.3%
32 16051 3.9%
64 14874
I find it pretty hard to beleive that 8k is precisely where a drop in
performance shows up unless there's some peculiar reason.
The only peculiar reason I can imagine is full page writes. If the
dbt2 workload is modifying already full pages then the full page
writes will always be just shy
Hi all,
A long long time ago (in 2004) I ran a series of tests surveying the
results of changing BLCKSZ when it was used for both the WAL logs and
the rest of the database system:
http://archives.postgresql.org/pgsql-hackers/2004-03/msg01194.php
Now more than 5 years later and now being able to