Thread: executing pgsql on Xeon-dual machine

executing pgsql on Xeon-dual machine

From
yutaka_inada@justsystem.co.jp
Date:
Hello,

I'm facing to a performance problem, when I run PostgreSQL 7.
2.1 + RedHat Linux 7.3 on a Xeon Dual CPU machine.

Compared with a P4 2.4GHz machine, pgbench shows 1/10 on the
Xeon machine (detail is describled below), while hdparm
shows 2 times faster.

Disabling HTT had no effects. PostgreSQL 7.2.3 is also slow
on the machine.

I'm grateful if any of you can give me an advice.

thank you.


data ---------------

[Hardware Profile]
DELL PowerEdge 2600
    CPU     Xeon 2GHz Dual
    Memory  2GB PC2100 ECC DDR266 SDRAM
    HDD     73GB 10,000rpm U320 SCSI

DELL PowerEdge 600SC (for comparison)
    CPU     Pentium 4 2.4GHz Single
    Memory  1GB 400MHz ECC DDR SDRAM
    HDD     80GB 7,200rpm EIDE

[pgbench result]

multiplicity  |  1  |  128
--------------+-----+------
PE 600        |259.2|178.2  [tps]
PE 2600       | 22.9| 34.8  [tps]

- wal_sync_method(wal_method.sh) fsync
- 200,000 data items

---
Yutaka Inada [Justsystem Corporation]

Re: executing pgsql on Xeon-dual machine

From
Tom Lane
Date:
yutaka_inada@justsystem.co.jp writes:
> Compared with a P4 2.4GHz machine, pgbench shows 1/10 on the
> Xeon machine (detail is describled below), while hdparm
> shows 2 times faster.

If you set fsync off, how do the pgbench results change?

> [Hardware Profile]
> DELL PowerEdge 2600
>     CPU     Xeon 2GHz Dual
>     Memory  2GB PC2100 ECC DDR266 SDRAM
>     HDD     73GB 10,000rpm U320 SCSI

> DELL PowerEdge 600SC (for comparison)
>     CPU     Pentium 4 2.4GHz Single
>     Memory  1GB 400MHz ECC DDR SDRAM
>     HDD     80GB 7,200rpm EIDE

I'm suspicious that the IDE drive may be configured to lie about write
completion.  If it reports write complete when it's really only buffered
the data in controller RAM, then you're effectively running with fsync
off on the PE600.

            regards, tom lane

Re: executing pgsql on Xeon-dual machine

From
yutaka_inada@justsystem.co.jp
Date:
Thank you, Tom,

Tom Lane <tgl@sss.pgh.pa.us>:
>
> If you set fsync off, how do the pgbench results change?

I'll try it.

---
Yutaka Inada [Justsystem Corporation]