Logical Volumes

[Alain St-Denis]

In article <72895 at sgi.sgi.com>, daveh at xtenk.asd.sgi.com (David A Higgen)
writes:
> 
> Suffering from a touch of multiple-post-itas, aren't we Alain? As I said
> in my last follow-up, I will try to answer specific questions.

Sorry about that.

> 
> No, no: this seems to be a widespread confusion. Growfs extends the
> *filesystem*, not the logical volume. A logical volume is extended by
> adding more devices to its specification in /etc/lvtab and rerunning
> mklv; you would do this *before* using growfs.

AH! I get it now... Thank you for the clarification (I'm not gonna say that
I knew, you wouldn't believe me :-) ...)

> I think there is some confusion because many people have heard the 
> terms "striped files" or "striped filesystems".
> 
> There are basically two ways you can do volume management & software
striping.
> 
> 1)  You can modify the filesystem to have knowledge of the physical devices
>     on which it resides, and add management policies to the filesystem 
>     itself to allow it to determine which physical disk a given file block
>     is on,
> 
> OR
> 
> 2)  You can build an intermediate layer, below the filesystem but above the
>     actual disk drivers, which makes multiple real disk devices look to
>     higher layers like a single device.
> 
> We have used the latter approach: it is more modular and allows volumes to
> be constructed from any type of physical disk without modifications to the
> filesystem code. Our logical volumes are *devices*, the filesystem knows
> nothing about them.

So, if I understand correctly, the logical volume layer decides where the
data blocks end up? If I have a filesystem that I want to "grow", will it
start using all the drives immediately or will it fill up the first one and
then start to use the next and so on?

And if I decide to use striping on a new logical volume, does the driver
see the physical disks as one disk and consider that, for example, a track
starts on the first and ends on the last disk. That would give faster
transfer rates but allows contention on the device, right? Or, are the data
blocks allocated based on whatever disk is not busy on the logical device?

What I want to know is that by using the logical volumes, I won't strangle
my system by making it think that it has only one fast disk instead of n
moderatly fast disks.

I think these are specific questions.
Feel free to criticize my limited knowledge of the topic (and my not so
bad but not so good english).
Thank you.
-- 
Alain St-Denis
Centre informatique de Dorval
Environnement Canada
astdenis at cid.aes.doe.CA
(514) 421-4697