Re: performance problem - 10.000 databases

W liście z czw, 06-11-2003, godz. 15:37, Jeff pisze:
> On 06 Nov 2003 15:21:03 +0100
> Marek Florianczyk <franki@tpi.pl> wrote:
>
>
> > fsync = false
>
> HOLD THE BOAT THERE BATMAN!
>
> I would *STRONGLY* advise not running with fsync=false in production as
> PG _CANNOT_ guaruntee data consistancy in the event of a hardware
> failure.  It would sure suck to have a power failure screw up your nice
> db for the users!

Sure I know, but with WAL it will make fsync every some? seconds, right?
Maybe users data, aren't so critical ;) it's not for bank, only for www
sites.
I will try with fsync=true also.

>
>
> > wal_buffers = 1024
>
> This also seems high. come to think about it- shared_buffers is also
> high.
>
> > commit_delay = 10000
>
> I could also read to data loss, but you'll get a speed increase on
> inserts.
>
> One of the best things you can do to increase insert speed is a nice,
> battery backed raid card with a pile of disks hanging off of it.

we will put 4 disks for /data directory ( raid1+0 ) so it will have
performance and fault tolerance, so it should be OK.

greetings
Marek