Re: x206-x225 - Mailing list pgsql-performance

From Daniel Blaisdell
Subject Re: x206-x225
Date
Msg-id f2af156d0603100627j39303849uc0e17d3ff468b0f@mail.gmail.com
Whole thread Raw
In response to Re: x206-x225  (Richard Huxton <dev@archonet.com>)
List pgsql-performance
The primary slow down is probably between your system bus from main memory to your disk storage. If you notice from your statistics that the select statements are very close. This is because all the data you need is already in system memory. The primary bottle neck is probably disk I/O. Scsi will always be faster than ATA. Scsi devices have dedicated hardware for getting data to and from the disc to the main system bus without requiring a trip through the CPU.

You may be able to speed up the ata disc by enabling DMA by using hdparm.

hdparm -d1 /dev/hda  (or whatever your device is)

-Daniel

-----BEGIN PGP SIGNATURE-----
Version: GnuPG v1.4.2 (GNU/Linux)

iD8DBQBEEYzX9SJ2nhowvKERAoiFAKCLR+7a7ReZ2mjjPjpONHLGIQD1SgCeNNON
V1kbyATIFVPWuf1W6Ji0IFg=
=5Msr
-----END PGP SIGNATURE-----

On 3/10/06, Richard Huxton <dev@archonet.com> wrote:
H.J. Sanders wrote:
> X206                                       IBM X226
> ----------------------                            -------------------
> processor                        Pentium 4 3.2
> Ghz                            Xeon 3.0 Ghz
> main memory                    1.25
> GB                                        4 GB
> discs                              2 x SCSI RAID1  10000RPM
> 1 x ATA 7200 RPM

Noting that the SCSI discs are on the *slower* machine.

> Time at X206            Time at X226
> --------------------        ------------------
> insert record (1 to 10000)          6 sec.    41 sec.
> select record (1 to 10000)          4          4
> delete record (1 to 10000)          6         41
>
>
> This is ofcourse a totally unexpected results (you should think off the
> opposite).

Your ATA disk is lying about disk caching being turned off. Assuming
each insert is in a separate transaction, then it's not going to do
10,000 / 6 = 1667 transactions/sec - that's faster than it's rotational
speed.

> Funny is that the select time is the same for both machines.

Because you're limited by the speed to read from RAM.

By the way - these sort of tests are pretty much meaningless in any
practical terms.
--
   Richard Huxton
   Archonet Ltd

---------------------------(end of broadcast)---------------------------
TIP 6: explain analyze is your friend

pgsql-performance by date:

Previous
From: Richard Huxton
Date:
Subject: Re: Query time
Next
From: "Hakan Kocaman"
Date:
Subject: Re: Hanging queries on dual CPU windows