Re: strange performance regression between 7.4 and 8.1

At 02:43 PM 3/2/2007, Alex Deucher wrote:
>On 3/2/07, Ron <rjpeace(at)earthlink(dot)net> wrote:
>>
>>...and I still think looking closely at the actual physical layout of
>>the tables in the SAN is likely to be worth it.
>
>How would I go about doing that?
>
>Alex

Hard for me to give specific advice when I don't know what SAN
product we are talking about nor what kind of HDs are in it nor how
those HDs are presently configured...

I quote you in an earlier post:
"The RAID groups on the SAN were set up for maximum capacity rather
than for performance. Using it for the databases just came up recently."

That implies to me that the SAN is more or less set up as a huge 105
HD (assuming this number is correct? We all know how "assume" is
spelled...) JBOD or RAID 5 (or 6, or 5*, or 6*) set.

=IF= that is true, tables are not being given dedicated RAID
groups. That implies that traditional lore like having pg_xlog on
dedicated spindles is being ignored.
Nor is the more general Best Practice of putting the most heavily
used tables onto dedicated spindles being followed.

In addition, the most space efficient RAID levels: 5* or 6*, are not
the best performing one (RAID 10 striping your mirrors)

In short, configuring a SAN for maximum capacity is exactly the wrong
thing to do if one is planning to use it in the best way to support
DB performance.

I assume (there's that word again...) that there is someone in your
organization who understands how the SAN is configured and administered.
You need to talk to them about these issues.

Cheers,
Ron