Re: Unexpected pgbench result

On Fri, Dec 20, 2013 at 7:10 AM, Shaun Thomas <sthomas(at)optionshouse(dot)com>wrote:

> On 12/19/2013 04:06 PM, Dave Johansen wrote:
>
> Right now, we're running a RAID 1 for pg_clog, pg_log and pg_xlog and
>> then a RAID 1+0 with 12 disks for the data. Would there be any benefit
>> to running a separate RAID 1+0 with a tablespace for the indexes?
>>
>
> Not really. PostgreSQL doesn't currently support parallel backend fetches,
> aggregation, or really anything. It's looking like 9.4 will get us a lot
> closer to that, but right now, everything is serial.
>
> Serial or not, separate backends will have separate read concerns, and
> PostgreSQL 9.2 and above *do* support index only scans. So theoretically,
> you might actually see some benefit there. If it were me and I had spindles
> available, I would just increase the overall size of the pool. It's a lot
> easier than managing multiple tablespaces.
>

Ok, that makes sense. Is there a benefit to having the WAL and logs on the
separate RAID 1? Or is just having them be part of the larger RAID 1+0 just
as good?