freebsd stable - Oct 2004 - question on vinum

Hello,

On our next file server, I want to make one large FTP area out of 4 drives (so 
that system sees them as one volume). Vinum appears to be exactly what I 
need.

I haven't worked with Vinum previously, but hear a lot about it. My question
is how to implement the above (unite four drives into single volume) using 
Vinum, and what will happen if let's say one drive fails in volume? Am I 
loosing the whole data, or I can just unplug the drive, tell vinum to use 
remaining drives with the data each drive holds? I'm not looking for fault 
tolerance solution.

From my understanding, in above scenario, Vinum will first fill up the first 
drive, then second, etc.

I have read the handbook articles, and I got general understanding of Vinum.
I'm particularly interested to know if I will still be able to use volume in
case of failed drive.
Some minimal configuration examples would be greatly appreciated!

Server will be running FreeBSD-4.10.

regards,
M.

On Tue, 2004-10-26 at 18:27, Mikhail P. wrote:
> I haven't worked with Vinum previously, but hear a lot about it. My
question
> is how to implement the above (unite four drives into single volume) using 
> Vinum, and what will happen if let's say one drive fails in volume? Am
I
> loosing the whole data, or I can just unplug the drive, tell vinum to use 
> remaining drives with the data each drive holds? I'm not looking for
fault
> tolerance solution.
Since you don't care about fault tolerance, you probably want to do
striping, also known as raid0.
> From my understanding, in above scenario, Vinum will first fill up the
first
> drive, then second, etc.
thats called concatenation, which is different than striping.  striping
balances the load across all the spindles
> I have read the handbook articles, and I got general understanding of
Vinum.
> I'm particularly interested to know if I will still be able to use
volume in
> case of failed drive.
If you want to do that, then you want raid5.  If either a concat or
stripe set looses a drive, the data will need to be restored.  
> Some minimal configuration examples would be greatly appreciated!
Read the following.  Really!
http://www.vinumvm.org/vinum/vinum.ps  
http://www.vinumvm.org/cfbsd/vinum.txt
Both of these have examples and will clear up your confusion about
concat vs stripe vs raid5.

concat is the easiest to add to, stripe has the best performance, raid5
trades write speed and n+1 drives for resilience.  raid10 gets back the
performance at the cost of 2*n drives 

Broken down:
volume - top level.  what the filesystem talks to.  mirroring is defined
at the volume level as is raid 10 (mirrored stripe plexes)
plex   - a virtual storage area made up of 1 or more subdisks for concat
2 or more for stripe, or 3 or more subdisks for raid 5.
subdisk - area delegated from a bsd partition
drive - the actual bsd partition (as in /dev/da1s1h)

generally, the order is as follows:
-fdisk the drives to be used so they have at least one bsd slice each. 
-use disklabel to edit the slice label so you have at least one
partition of type vinum (that isn't the C partition)
-in an editor, create the configuration
drives
volume
  plex
    sd

when you define the subdisk, don't use the whole drive.  Leave at least
64 blocks unused.

use the file you created as input to vinum
vinum create -v -f config

Or you can cheat and just say, 
"vinum stripe -n volname /dev/ad0s1h /dev/ad1s1h /dev/ad2s1h
/dev/ad3s1h" (should be all on one line)

Raid5 plexs have to be init'ed 

newfs -v /dev/vinum/volname
mount /dev/vinum/volname /mnt

hopefully I haven't made you're understanding worse

Marko Raiha wrote:
>Hi,
>
>On Wed, Oct 27, 2004 at 09:10:39AM -0600, secmgr wrote:
>  
>
>>The reason is empirically derived.  When I created a 7 disk raid 5 set 
>>using "len 0" or all the space available, the raid set would
be corrupt
>>after initializing.  Every time.  When I reserved back  that extra 
>>space, no corruption.
>>(freebsd 4.10-p3)  There was a thread on this a few days ago.
>>    
>>
>
>Thanks about pointing to this another thread. I read the thread and one
>thing is still a bit unclear. Do you mean by leaving 64 blocks unused to
>leave it unused at the beginning of disk before vinum partition (when
>doing partition to vinum by disklabeling freebsd slice) like this:
>
># /dev/aacd2s1:
>8 partitions:
>#        size   offset    fstype   [fsize bsize bps/cpg]
>  c: 143347932        0    unused        0     0 
>  h: 143347868       64    vinum
>
>or somewhere in subdisk as you said? I guess that inadvertently not
>right terms were used ("subdisk" instead of "disk"),
because terminology
>with vinum is a bit complicated.
>
>All documentations I have found says that only 16 block should be left
>unused at the beginning of the disk like this:
>
># /dev/aacd2s1:
>8 partitions:
>#        size   offset    fstype   [fsize bsize bps/cpg]
>  c: 143347932        0    unused        0     0
>  h: 143347916       16    vinum
>
>Regards,
>  
>
That probably would have worked too.  My mistake was to clone the c 
partition entry, call it h, label it vinum, and use that