On Fri, 22 Feb 2008, Tom Lane wrote:
> Counts are useless here, we need to see the sequence of write locations
> to find out if there's a lot of nonconsecutive writes happening.
How were you planning to analyze the strace output to quantify that? I
would think you'd need to parse the file to grab those locations then
generate statistics about seeking from them. Those strace files will be
huge and I'd expect Joshua would need to do something like that on his
side before the results will be short enough to share.
> it's quite possible that there's some interaction between the bgwriter
> and the backend doing COPY
It might be worthwhile to turn off the background writer for these runs,
both to see if it changes anything and to simplify the output.
--
* Greg Smith gsmith@gregsmith.com http://www.gregsmith.com Baltimore, MD