Re: Tuning PostgreSQL

On Tue, 2003-07-29 at 11:18, scott.marlowe wrote:
> On 29 Jul 2003, Ron Johnson wrote:
>
> > On Tue, 2003-07-29 at 10:14, Vivek Khera wrote:
> > > >>>>> "GS" == Greg Stark <gsstark@mit.edu> writes:
> > >
> > > GS> "scott.marlowe" <scott.marlowe@ihs.com> writes:
> > >
> > > GS> But you have to actually test your setup in practice to see if it
> > > GS> hurts. A big data warehousing system will be faster under RAID5
> > > GS> than under RAID1+0 because of the extra disks in the
> > > GS> stripeset. The more disks in the stripeset the more bandwidth you
> > > GS> get.
> > >
> > > Anyone have ideas on 14 spindles?  I just ordered a disk subsystem
> > > with 14 high speed (U320 15kRPM) SCSI disks to hook up with a dell
> > > PERC3/DC controller (only 128MB cache, though).
> >
> > 14 drives on one SCSI card, eh?  I'd be worried about saturating
> > the bus.
>
> I'm pretty sure those PERCs are based on the megaraid cards, which can
> handle 3 or 4 channels each...

Each with 14 devices?  If so, isn't that a concentrated point of
failure, even if the channels are 1/2 full?

> > Maybe it's an old rule of thumb, but I would fill a SCSI chain
> > more than half full.
>
> It's an old rule of thumb, but it still applies, it just takes more drives
> to saturate the channel.  Figure ~ 30 to 50 MBytes a second per drive, on
> a U320 port it would take 10 drives to saturate it, and considering random
> accesses will be much slower than the max ~30 megs a second off the
> platter rate, it might take more than the max 14 drives to saturate U320.

Ok.  You'd still saturate the 133MB/s PCI bus at 133/30 = 4.4 drives.

--
+-----------------------------------------------------------------+
| Ron Johnson, Jr.        Home: ron.l.johnson@cox.net             |
| Jefferson, LA  USA                                              |
|                                                                 |
| "I'm not a vegetarian because I love animals, I'm a vegetarian  |
|  because I hate vegetables!"                                    |
|    unknown                                                      |
+-----------------------------------------------------------------+