Greg Smith wrote: > On Thu, 22 Jan 2009, Alvaro Herrera wrote: > >> Also, I think you should set the "scale" in the prepare step (-i) at >> least as high as the number of clients you're going to use. (I dimly >> recall some recent development in this area that might mean I'm wrong.) > > The idea behind that maxim (clients>=scale) is that locking on the > smaller tables will bottleneck resuls if you don't follow that advice. > It's a bit messier than that though. Increasing the scale will also > make the database larger, and once it gets bigger than available RAM > your results are going to dive hard because of that, more so than the > locking would have held you back. > > All kind of irrelevant for Ibrahim's case, because if you're not getting > more than 50MB/s out of your disks the pgbench results are kind of moot > anyway--there's a larger problem to sort out first. > > -- > * Greg Smith gsmith@xxxxxxxxxxxxx http://www.gregsmith.com Baltimore, MD > IIRC this is a FreeBSD system, not Linux. Could there be some filesystem performance issue here? I know zero about FreeBSD filesystems. Also, is there a separate driver machine you can use to run pgbench? The pgbench client uses resources, which could lower your throughput. -- M. Edward (Ed) Borasky I've never met a happy clam. In fact, most of them were pretty steamed. -- Sent via pgsql-performance mailing list (pgsql-performance@xxxxxxxxxxxxxx) To make changes to your subscription: http://www.postgresql.org/mailpref/pgsql-performance
