On 11/08/2011 02:49 PM, Ondrej Ivanič wrote:
> Usually CPU is not bottleneck but I it was when I put Pustgres on
> FusionIO. The problem is that PG spreads reads too much . iostat
> reports very low drive utilisation and very low queue size.
"Spreads reads too much" ?
Are you saying there's too much random I/O? Is it possible it'd benefit
from a column store?
When you're using Greenplum are you using "Polymorphic Data Storage"
column storage "WITH (orientation=column)" ?
Or is the performance different just in better utilisation of the
hardware under Greenplum?
> I've played with effective_io_concurrency (went thru entire range: 1,
> 2, 5, 10, 20, 50, 100, 200, 500, 1000) but nothing improved.
Hm, ok. Scratch that then.
> Is there
> a way to get PG backed IO stats using stock CentOS (5.7) kernel and
> tools? (I can't change my env easily)
Dunno; check postgresql high performance (book), the manual, etc. Useful
tools are the pg_stat_ tables, "vmstat", "iostat", "iotop", etc.
>
>> I'm not at all surprised by that. PostgreSQL couldn't use the full resources
>> of your system when it was expressed as just one query.
> This is very interesting area to work in but my lack of C/C++ and PG
> internals puts me out of the game :)
That's a cop-out! I say that as someone who _used_ to have no C
knowledge, but learned it to get things done on code I use.
That said, Pg's codebase isn't exactly trivial :S and trying to get
involved in major re-engineering like parallelisation isn't going to be
practical when you're just getting started.
--
Craig Ringer