[Bizgres-general] Performance - WAL bypass + parse

[Mark Kirkwood]

Mark Kirkwood wrote:

> 
> My money is on the IO *not* being the bottleneck - I will see if I can 
> get these profiles going!
> 
>

Hmmm - I suspect that the profiles are telling me that my money is going 
to Luke....

The flat graphs for vanilla vs fast show sampled times of 1113s vs 792s 
- which suggests that about 50% of the cpu time is being saved....

Also, the elapsed times show 50% improvement for the fast case. This 
does suggest that the slowing effect of the profiling is enabling the IO 
or memory subsystem to keep up!

You can check them out at:

http://homepages.paradise.net.nz/markir/download/postgres-copy-profiles.tar.gz

 > That concurrent read/write perf is terrible.
 >

Yes - I guess it is hitting the crap seek times of the ATA disks.

 > When you compiled, you did so with optimization, right?
 >

yes - the usual -O2


 > Is this a 2 CPU machine?  Can you try a load from a networked client
 > using backslash copy?
 >

Will do - interesting suggestion!

Cheers

Mark