Re: Poor performance on seq scan

From: Dave Cramer <pg(at)fastcrypt(dot)com>
To: Laszlo Nagy <gandalf(at)designaproduct(dot)biz>
Cc: pgsql-performance(at)postgresql(dot)org
Subject: Re: Poor performance on seq scan
Date: 2006-09-13 12:22:37
Message-ID: 1BD8A62B-DE69-4EFA-9EB6-D11B7CBDDA9E@fastcrypt.com
Views: Raw Message | Whole Thread | Download mbox | Resend email
Thread:
Lists: pgsql-performance


On 13-Sep-06, at 6:16 AM, Laszlo Nagy wrote:

>
>> I have had extremely bad performance historically with onboard
>> SATA chipsets
>> on Linux. The one exception has been with the Intel based
>> chipsets (not the
>> CPU, the I/O chipset).
>>
> This board has Intel chipset. I cannot remember the exact type but
> it was not in the low end category.
> dmesg says:
>
> <Intel ICH7 SATA300 controller>
> kernel: ad4: 152626MB <SAMSUNG HD160JJ ZM100-33> at ata2-master
> SATA150
> kernel: ad4: 152627MB <SAMSUNG HD160JJ ZM100-33> at ata3-master
> SATA150
>
>> It is very likely that you are having problems with the driver for
>> the
>> chipset.
>>
>> Are you running RAID1 in hardware? If so, turn it off and see
>> what the
>> performance is. The onboard hardware RAID is worse than useless, it
>> actually slows the I/O down.
>>
> I'm using software raid, namely gmirror:
>
> GEOM_MIRROR: Device gm0 created (id=2574033628).
> GEOM_MIRROR: Device gm0: provider ad4 detected.
> GEOM_MIRROR: Device gm0: provider ad6 detected.
> GEOM_MIRROR: Device gm0: provider ad4 activated.
> GEOM_MIRROR: Device gm0: provider ad6 activated.
>
> #gmirror list
> Geom name: gm0
> State: COMPLETE
> Components: 2
> Balance: round-robin
> Slice: 4096
> Flags: NONE
> GenID: 0
> SyncID: 1
> ID: 2574033628
> Providers:
> 1. Name: mirror/gm0
> Mediasize: 160040803328 (149G)
> Sectorsize: 512
> Mode: r5w5e6
> Consumers:
> 1. Name: ad4
> Mediasize: 160040803840 (149G)
> Sectorsize: 512
> Mode: r1w1e1
> State: ACTIVE
> Priority: 0
> Flags: DIRTY
> GenID: 0
> SyncID: 1
> ID: 1153981856
> 2. Name: ad6
> Mediasize: 160041885696 (149G)
> Sectorsize: 512
> Mode: r1w1e1
> State: ACTIVE
> Priority: 0
> Flags: DIRTY
> GenID: 0
> SyncID: 1
> ID: 3520427571
>
>
> I tried to do:
>
> #sysctl vfs.read_max=32
> vfs.read_max: 6 -> 32
>
> but I could not reach better disk read performance.
>
> Thank you for your suggestions. Looks like I need to buy SCSI disks.

Well before you go do that try the areca SATA raid card
>
> Regards,
>
> Laszlo
>
>
> ---------------------------(end of
> broadcast)---------------------------
> TIP 5: don't forget to increase your free space map settings
>

In response to

Responses

Browse pgsql-performance by date

  From Date Subject
Next Message yoav x 2006-09-13 12:24:14 sql-bench
Previous Message Laszlo Nagy 2006-09-13 10:16:36 Re: Poor performance on seq scan