On Wed, 12 Jul 2006 10:16:40 -0600
"Koth, Christian (DWBI)" <Christian.Koth@smiths-heimann.com> wrote:
> I have noticed a strange performance behaviour using a commit statement on two different machines. On one of the
machinesthe commit is many times faster than on the other machine which has faster hardware. Server and client are
runningalways on the same machine.
>
> Server version (same on both machines): PostgreSQL 8.1.3. (same binaries as well)
>
> PC1:
> ----
> Pentium 4 (2.8 GHz)
> 1GB RAM
> IDE-HDD (approx. 50 MB/s rw), fs: ext3
> Mandrake Linux: Kernel 2.4.22
>
>
> PC2:
> ----
> Pentium 4 (3.0 GHz)
> 2GB RAM
> SCSI-HDD (approx. 65 MB/s rw), fs: ext3
> Mandrake Linux: Kernel 2.4.32
>
>
> Both installations of the database have the same configuration, different from default are only the following
settingson both machines:
>
> shared_buffers = 20000
> listen_addresses = '*'
> max_stack_depth = 4096
>
>
> pgbench gives me the following results:
> PC1:
> ----
> transaction type: TPC-B (sort of)
> scaling factor: 1
> number of clients: 1
> number of transactions per client: 10
> number of transactions actually processed: 10/10
> tps = 269.905533 (including connections establishing)
> tps = 293.625393 (excluding connections establishing)
>
> PC2:
> ----
> transaction type: TPC-B (sort of)
> scaling factor: 1
> number of clients: 1
> number of transactions per client: 10
> number of transactions actually processed: 10/10
> tps = 46.061935 (including connections establishing)
> tps = 46.519634 (excluding connections establishing)
I'm not sure 10 transactions is enough of a test. You could just be
seeing the result of your IDE drive lying to you about actually writing
your data. There may be other considerations but I would start with
checking with 10,000 or 100,000 transactions to overcome the driver
buffering.
--
D'Arcy J.M. Cain <darcy@druid.net> | Democracy is three wolves
http://www.druid.net/darcy/ | and a sheep voting on
+1 416 425 1212 (DoD#0082) (eNTP) | what's for dinner.
