UNIX for Advanced & Expert Users

Hello all. My question is can you hang a filesystem any where in a path?

Working Example:

/abc/example/new_filesystem

I know that /abc in my example above would be sitting on its own hd, so is it possible to hang a completely new filesystem (new_filesystem) couple of hierarchies down (as in example above, that is under /abc/example..) ?

2nd question:
Is there any I/O benefits in the following example, which one you would chose (this filesystem will have high I/O sequential reads and writes, about 5% random writes and reads)

/xyz/highio/special_temp or /special_temp ?

Your opinion is deeply appreciated.

* All on AIX 64
** All supposed to be separate and unique filesystems.

In the beginning I did mount FS into other FS too. It is no good idea at all since if you have a Cluster of any kind, let's say HA/CMP for example, there can be problems in the mount order when a Resource Group Takeover is occuring cause one node crashed.
You can care about that manually, but when building clusters it is best to have all resources as plain as possible.

Since you have with AIX a very capable LVM, you can easily put more PVs into the VG if it is low on free space even while the system is up and running at production times. Also you can easy expand filesystems.
For performance considerations, if the LV is on different disks, yes, it will give more performance basically. It mostly depends what kind of disks are underneath etc.

You mean mount...
Yes you can but you must pay attention that the mount point is a mount point ( empty directory...) otherwise you will mask the content of the subdirectories underneath the new filesystem... a classical case: /opt
/opt will be on the main OS disk but you may not like to have all oracle stuff on tha same disk...
you will create for oracle software ( not the DB!) a filesystem and mount it on :
/opt/oracle/product
I usually do one per version, so you would have
/opt/oracle/product/9i
/opt/oracle/product/10g
etc...

Are you on SAN?
You could add a couple of PVs to your VG then use LVM stripping to augment your IO throughput (I am not on AIX on the moment but I remember there is something of the sort we used to call the stripping of the poor in LVM configuration...)

Hi VBE, yes we are on SAN.

VBE, so in plain english, it is OK to have a FS as xyz/my/example/newfs (where newfs is a new file system) as you showed in your example. The 2nd item, was which is better, in terms of performance a FS in a path or mounted off of /. This is for temp space (high hits)

/example/my/newfs_temp or if we mount it /newfs_temp ?

* Sorry I am not well versed on the SAN side of things

I forgot to outline zaxxon's remark: mounting filsystems on filesystems can bring up issues like if the mounting order isnt respected, you have weard behaviour:
We have
/opt/oracle
/opt/oracle/product/ #directory for all the versions mountpoints
/opt/oracle/product/8.1.6 #just for the example.
It happened once on a AIX that /opt/oracle/product/8.1.6 got mounted first
then
/opt/oracle ...
The dbas were in panic because they thought they had lost (a bad rm ????) their software
They started to restore in /opt/oracle all the missing product, at the time I came in.
And after looking at df I told them to stop, umounted /opt/oracle, /opt/oracle/product/8.1.6
Then mounted in correct sequence.
All was there...
This is to say beware of people that can create the path on /
The above situation should never have occured (well not that way...): If no one had created /opt/oracle/product/8.1.6 in / that filesystem could not mount till /opt/oracle was...

About your second remark
Usually SAN means great disks subsystems with big cache, check!
If so external is advisable

I tend do leave the strict minimum in rootvg: The entire OS..and mirror that
The rest goes in diverse VG
Its easier for OS backup (mksysb) and upgrade...