Thread: asynchronous commit feature

asynchronous commit feature

From
"Merlin Moncure"
Date:
I'm testing the new asynch commit feature on various raid
configurations and my early findings is that it reduces the impact of
keeping wal and data on the same volume.  I have 10 disks to play
with, and am finding that it's faster to do a 10 drive raid 10 rather
than 8 drive raid 10 + two drive wal.

anybody curious about the results, feel free to drop a line.  I think
this will be a popular feature.

merlin

Re: asynchronous commit feature

From
Decibel!
Date:
On Thu, Aug 23, 2007 at 09:09:00AM -0400, Merlin Moncure wrote:
> I'm testing the new asynch commit feature on various raid
> configurations and my early findings is that it reduces the impact of
> keeping wal and data on the same volume.  I have 10 disks to play
> with, and am finding that it's faster to do a 10 drive raid 10 rather
> than 8 drive raid 10 + two drive wal.
>
> anybody curious about the results, feel free to drop a line.  I think
> this will be a popular feature.

With or without a write cache on the RAID controller? I suspect that for
many systems, a write-caching controller will be very similar in
performance to async commit.
--
Decibel!, aka Jim Nasby                        decibel@decibel.org
EnterpriseDB      http://enterprisedb.com      512.569.9461 (cell)

Attachment

Re: asynchronous commit feature

From
"Merlin Moncure"
Date:
On 8/27/07, Decibel! <decibel@decibel.org> wrote:
> On Thu, Aug 23, 2007 at 09:09:00AM -0400, Merlin Moncure wrote:
> > I'm testing the new asynch commit feature on various raid
> > configurations and my early findings is that it reduces the impact of
> > keeping wal and data on the same volume.  I have 10 disks to play
> > with, and am finding that it's faster to do a 10 drive raid 10 rather
> > than 8 drive raid 10 + two drive wal.
> >
> > anybody curious about the results, feel free to drop a line.  I think
> > this will be a popular feature.
>
> With or without a write cache on the RAID controller? I suspect that for
> many systems, a write-caching controller will be very similar in
> performance to async commit.

I usually only work with mid to high end hardware.

The platform I'm testing on is:
10x146gb 15k rpm sas in Dell md1000
2xperc 5/e in active/active (5 drives each controller)
2x146gb 15krpm sas on the backplane perc 5/i (for o/s, wal)

in my experience, even with a high end raid controller moving the wal
offline is helpful in high activity systems, especially during
checkpoints.

merlin