Thread: recommended way of separating data from indexes

recommended way of separating data from indexes

From
"Anton Melser"
Date:
Hi,
I looked for quite a while but must have been looking in the wrong place...
I have a 6 disk (x146GB) system running Centos 5.1 - RAID 1 (2 disks)
and RAID 1+0 (4 disks). This seemed to be the recommended way of doing
it but I may have been looking in the wrong place. The system is
replicated using log shipping, so we have the archive command
activated.
What are the recommendations for this kind of system? The system is on
the RAID 1 disk. My idea was to separate only the data onto the RAID
1+0 and leave the logs and archive and rest on the RAID 1. Is this the
way to go? What is the recommended way of doing this? Just using
symbolic links? Help!
Cheers
Anton

--
echo '16i[q]sa[ln0=aln100%Pln100/snlbx]sbA0D4D465452snlbxq' | dc
This will help you for 99.9% of your problems ...

Re: recommended way of separating data from indexes

From
Richard Huxton
Date:
Anton Melser wrote:
> Hi,
> I looked for quite a while but must have been looking in the wrong place...
> I have a 6 disk (x146GB) system running Centos 5.1 - RAID 1 (2 disks)
> and RAID 1+0 (4 disks). This seemed to be the recommended way of doing
> it but I may have been looking in the wrong place. The system is
> replicated using log shipping, so we have the archive command
> activated.
> What are the recommendations for this kind of system?

That's going to depend on your disk activity, which will depend on usage
patterns.

 > The system is on
> the RAID 1 disk. My idea was to separate only the data onto the RAID
> 1+0 and leave the logs and archive and rest on the RAID 1. Is this the
> way to go? What is the recommended way of doing this? Just using
> symbolic links? Help!

Read the tablespaces chapter of the manuals.

If you've got a fair amount of writing then you'll want to make sure the
WAL can be written to as quickly as possibly. If you've got other write
activity occurring on your system disks (e.g. lots of system-log
activity, mailserver etc) then that's probably not the best place for
the WAL. On the other hand, database reads+writes will interfere if you
put everything on the RAID 10.

So - it will depend.

--
   Richard Huxton
   Archonet Ltd

Re: recommended way of separating data from indexes

From
"Anton Melser"
Date:
Hi,

>> I have a 6 disk (x146GB) system running Centos 5.1 - RAID 1 (2 disks)
>> and RAID 1+0 (4 disks). This seemed to be the recommended way of doing
>> it but I may have been looking in the wrong place. The system is
>> replicated using log shipping, so we have the archive command
>> activated.
>> What are the recommendations for this kind of system?
>
> That's going to depend on your disk activity, which will depend on usage
> patterns.

It's a website, with probably around 98% read to 2% write, and the
writes only to a few tables.

>> The system is on
>>
>> the RAID 1 disk. My idea was to separate only the data onto the RAID
>> 1+0 and leave the logs and archive and rest on the RAID 1. Is this the
>> way to go? What is the recommended way of doing this? Just using
>> symbolic links? Help!
>
> Read the tablespaces chapter of the manuals.
>
> If you've got a fair amount of writing then you'll want to make sure the WAL
> can be written to as quickly as possibly. If you've got other write activity
> occurring on your system disks (e.g. lots of system-log activity, mailserver
> etc) then that's probably not the best place for the WAL. On the other hand,
> database reads+writes will interfere if you put everything on the RAID 10.
>
> So - it will depend.

The machine is only doing this one DB, with the odd email being sent
by postfix (20 p/d, so nothing) and for the moment we aren't doing any
real logging apart from postgres internal (so not even activity logs
or query logs, etc).

So given that disk usage is pretty much 100% pgsql, and it's mainly
read, does my architecture stand up? And thanks, I'll have a read of
the tablespaces chapter - I looked pretty much everywhere but there!
Cheers
Anton

Re: recommended way of separating data from indexes

From
Richard Huxton
Date:
Anton Melser wrote:
>
> It's a website, with probably around 98% read to 2% write, and the
> writes only to a few tables.

> The machine is only doing this one DB, with the odd email being sent
> by postfix (20 p/d, so nothing) and for the moment we aren't doing any
> real logging apart from postgres internal (so not even activity logs
> or query logs, etc).
>
> So given that disk usage is pretty much 100% pgsql, and it's mainly
> read, does my architecture stand up? And thanks, I'll have a read of
> the tablespaces chapter - I looked pretty much everywhere but there!

With a small number of writes, it probably doesn't matter too much.

If you symlink the PGDATA directory, or have it in a mount you can move
then you can always change your mind later.

--
   Richard Huxton
   Archonet Ltd