Thread: WALL on controller without battery?

WALL on controller without battery?

From
Francisco Reyes
Date:
Setting spec for a postgresql server.
The hard drive distribution is going to be
8 x 750GB Seagate on a 3ware 9650SE RAID 6
2 x 160GB Seagate on a 3ware 2 port

The question is, would I be better off putting WAL on the second, OS,
controller or in the 8 port controller? Specially since the 2 port will not
have battery (3ware does not have 2 ports with battery).
The two port controller is primary for the operating system, but I was
wondering if there would be any benefit to putting WAL on that 2 port
controller.

The machine will have 8Gb of RAM and will be running Postgresql 8.2 on
FreeBSD 6.2 Stable.

During peak operation there will be about 5 to 20 updates per second with a
handfull of reads.

Re: WALL on controller without battery?

From
Andrew Sullivan
Date:
On Wed, Jul 11, 2007 at 10:48:04AM -0400, Francisco Reyes wrote:
> The question is, would I be better off putting WAL on the second, OS,
> controller or in the 8 port controller? Specially since the 2 port will not
> have battery (3ware does not have 2 ports with battery).

Put the WAL where the battery is.  Even if it's slower (and I don't
know whether it will be), I assume that having the right data more
slowly is better than maybe not having the data at all, quickly.

A

--
Andrew Sullivan  | ajs@crankycanuck.ca
The plural of anecdote is not data.
        --Roger Brinner

Re: WALL on controller without battery?

From
Alan Hodgson
Date:
On Wednesday 11 July 2007 08:36, Andrew Sullivan <ajs@crankycanuck.ca>
wrote:
> Put the WAL where the battery is.  Even if it's slower (and I don't
> know whether it will be), I assume that having the right data more
> slowly is better than maybe not having the data at all, quickly.
>

Presumably he'll have the 2-port configured for write-through operation.

I would spring for a 4-port with a BBU, though, and then put the WAL on the
drives with the OS.

--
"Bad laws are the worst sort of tyranny." -- Edmund Burke (1729-1797)


Re: WALL on controller without battery?

From
Greg Smith
Date:
On Wed, 11 Jul 2007, Alan Hodgson wrote:

> Presumably he'll have the 2-port configured for write-through operation.

This is the real key to his question.  In order to get acceptable
performance for the operating system, Francisco may very well need the OS
disks to be configured in write-back mode.  If that's the case, then he
can't put the WAL there; it has to go onto the array with the BBU.

> I would spring for a 4-port with a BBU, though, and then put the WAL on the
> drives with the OS.

This is certainly worth considering.  When putting multiple RAID
controllers into a system, I always try to keep them of a similar grade
because it improves the possibility of data recovery in case of a
controller failure.  For example, if he had a 4-port with BBU and an
8-port with BBU, the 8-port could be split into two 4-disk RAID-6 volumes,
and then in an emergency or for troubleshooting isolation you could always
get any data you needed off any 4-disk set with either controller.  The
little 2-disk unit is providing no such redundancy.

--
* Greg Smith gsmith@gregsmith.com http://www.gregsmith.com Baltimore, MD

Re: WALL on controller without battery?

From
Francisco Reyes
Date:
Alan Hodgson writes:

> I would spring for a 4-port with a BBU, though, and then put the WAL on the
> drives with the OS.

The machine is already over budget. :-(
I will check the price difference but unlikely I will get approval.

Re: WALL on controller without battery?

From
"Joshua D. Drake"
Date:
Francisco Reyes wrote:
> Alan Hodgson writes:
>
>> I would spring for a 4-port with a BBU, though, and then put the WAL
>> on the drives with the OS.
>
> The machine is already over budget. :-(
> I will check the price difference but unlikely I will get approval.

Without a BBU you are guaranteed at some point to have catastrophic
failure unless you turn off write cache, which would then destroy your
performance.

Joshua D. Drake


>
> ---------------------------(end of broadcast)---------------------------
> TIP 7: You can help support the PostgreSQL project by donating at
>
>                http://www.postgresql.org/about/donate
>


--

       === The PostgreSQL Company: Command Prompt, Inc. ===
Sales/Support: +1.503.667.4564 || 24x7/Emergency: +1.800.492.2240
Providing the most comprehensive  PostgreSQL solutions since 1997
              http://www.commandprompt.com/

Donate to the PostgreSQL Project: http://www.postgresql.org/about/donate
PostgreSQL Replication: http://www.commandprompt.com/products/


Re: WALL on controller without battery?

From
Francisco Reyes
Date:
Joshua D. Drake writes:

> Without a BBU you are guaranteed at some point to have catastrophic
> failure unless you turn off write cache, which would then destroy your
> performance.

I am re-working the specs of the machine to try and get a 4port 3ware to
have the battery backup.

Re: WALL on controller without battery?

From
Greg Smith
Date:
On Wed, 11 Jul 2007, Francisco Reyes wrote:

> I am re-working the specs of the machine to try and get a 4port 3ware to have
> the battery backup.

That's really not necessary, it just would be better (and obviously more
expensive).  The warnings you've been getting here have been to let you
know that you absolutely can't put the WAL on the controller with the OS
disks attached without making compromises you probably won't be happy
with.

> During peak operation there will be about 5 to 20 updates per second
> with a handfull of reads.

There really is no reason you need to be concerned about WAL from a
performance perspective if this is your expected workload.  If you're
working with a tight budget, the original design you had was perfectly
fine.  Just use all the disks on the big controller as a large volume, put
both the database and the WAL on there, and don't even bother trying to
separate out the WAL.  If you expected hundreds of updates per second,
that's where you need to start thinking about a separate WAL disk, and
even then with 8 disks to spread the load out and a good caching
controller you might still be fine.

--
* Greg Smith gsmith@gregsmith.com http://www.gregsmith.com Baltimore, MD

Re: WALL on controller without battery?

From
Francisco Reyes
Date:
Greg Smith writes:

>> During peak operation there will be about 5 to 20 updates per second
>> with a handfull of reads.
>
> There really is no reason you need to be concerned about WAL from a
> performance perspective if this is your expected workload.


I was able to get the second controller with battery backup.
This machine is the backup so if the primary fails it would get higher
volumes.

It is also easier to throw more work at a good machine than to find myself
 with an underperformer.

> both the database and the WAL on there, and don't even bother trying to
> separate out the WAL.

Thanks for the feedback.
I wish there was a place with hardware guide where people could get feedback
like the one you gave me. In particular actual numbers like x to y number of
transactions per second you don't need WAL no separate disk.. etc..