create lvol with sc10 parameters instead of FC60 for a FC60 disk system

[John_912]

Dear all,
In a series of FC60 disks I wanted to create a new logicalvolume to one of the volume groups (in these disks a database is kept). Instead of giving the following command:lvcreate -L 2048 -i 12 -I 4 -n which is for FC60 disks, I gave the command lvcreate –D y –s g –m 1 -n -L 2048 which is for sc10 disks. The lvol was created successfully I continued the procedure and I added space for my database. What impact will the mistakenly given command will have on my system?
Thank you in ad

[Robert-Jan Goossens]

Hi John,

It is a bit difficult to read your message, but it should have no consequences to your system if you do not have data inside you new logical volume. Lvremove the newly created volume and recreate is in the good volume group.

Hope this helps,
Robert-Jan

[John_912]

Thanx, but already significant data are written. let me rephrase the question:
In my system I have FC60 disks,
now as far as I can underastand FC60 system of disks handles the disks differently than an SC10 system, that is why it was recommened whenever a creation of lvol is needed to run lvcreate with the following parameters:
"lvcreate -L 2048 -i 12 -I 4 ..."

Instead I mistakenly gave the command with parameters for SC10 disk system meaning:
"lvcreate -D y -s g -m 1 .."

To remove it unfortunatelly is not an option.

[Robert-Jan Goossens]

How about copying the database to a temporary file system ?

Close the database, use cpio/tar to create a copy, remove the volume, recreate the volume, and copy it back.

[John_912]

I cannot do this either, the system is real-time.For now I can read and write my data from the new lvol. I was informed that I might have a problem if the disk fails, it will not be able to do a proper build. But from what I can understand by your sayings, the only way is to get rid of the lvol. I was hoping to bypass this, because I know the procedure to do it but it is painfull and comes with various consequences.

[Robert-Jan Goossens]

Lets put it this way, you can not change the options you made with lvcreate. Take some maintenance time and recreate the volume.

Best regards,
Robert-Jan

[John_912]

Thank you. I will do so.

[Florian Heigl (new acc)]

There should be nothing worse than possible lower performance. Is the FC60 configured as a RAID5 or RAID1? Then I'd do a lvreduce -m 0 to remove the mirror copies from LVM.

[TwoProc · Accepted solution]

All you did is use larger stripes in a mirror that you didn't need.

A) you created stripes based on the size of your PE's - I use this all the time - not a bad thing... it's just that these stripes are much bigger then the stripes that originally wanted.

B) you created a mirror - which you probably didn't want since you probably wanted to let the FC60 do it for you.

Use lvdisplay to see if you have a mirror.
Use lvreduce to get rid of the mirror. If you're not sure how to do it and detect this - go play with a test box, and preferrably create an lvol using the same method - just undo it to show yourself you can do it "live" without hurting anything.

If you get rid of the mirror you'll just have a different kind of stripe size - which, depending on your application may not hurt you, and has a real possibility of even helping your performance.

(See documents on Metalink on SAME, as well as white papers from HP on striping to see why larger distributed stripes can really help your performance).

Right now the only thing that probably hurting you is a set of mirrored writes that you don't need. You can certainly break a mirror "hot" without taking a system down. That being said - please make sure and do it correctly so as not mess it up (I know, thanks for the obvious...) .

Suggestions
a) call HP for help, they can get you through the process of unmirroring live - no problem.
b) use one of your old boxes (even it doesn't have an FC60 on it) and play around with the whole idea - once you're comfortable with it - you'll see that this is really a simple thing to deal with.

P.S. I run several systems with direct attached SC10s ( I pulled the FC60's years ago) now, and I use distributed stripes (as you just mistakenly did) to try to get a bit more performance out of them - but to tell you the truth on the FC60's (only) the differnce in performance that I could actually measure was minimal... so, don't get too worked up about using distributed stripes vs regular stripes - in my experience it hardly mattered in my throughput tests. I still run distributed stripes just because it gave a little bit of performance boost on some types of expensive reads.

[Bob_Vance]

The difference between your two different 'lvcreate' commands is this:

... "-i" ... == let LVM do "striping". LVM will lay out the LVOL across more than one PV using a small stripe size that is smaller than a PE (Physical Extent), which is the normal allocation unit in LVM (4MB or 8MB, etc).

... "-D y" ... == use a PE-based striping -- i.e., stripe across more than one PV, but use a PE as the stripe.

So, theoretically, "-i" has the potential for slightly better striping performance. (But this is not necessarily the case.)

Why two different commands?
Because -i is not allowed with *LVM mirroring* -- just a LVM deficiency.
If you want to LVM mirror *and* stripe (a poor man's RAID1+0), then HP LVM requires the "-D y" -- not "-i".

Presumably, the FC60 is already doing the mirroring (RAID 1), so LVM Mirror is not required when using the FC60 LUNs. Note: if the FC60 is using RAID5 or RAID1+0, then striping isn't necessary either, here.

As said by the others above, the impact of your command is that you created an *unnecessary* LVM mirror for the FC60 LVOL.
This is wasting space = LVOL size.
You can simply remove the mirror.
At this point you will only be using the "-D y" striping, which, as said above, has *potentially* very slightly less performance; but might even be better, as J.J. mentioned.

tks
bv