Summary: mirrored LSM volumes - time to create

From: <anthony.miller_at_vf.vodafone.co.uk>
Date: Fri, 05 Feb 1999 11:05:53 +0000

Many thanks to those who responded:

Hank Lee <hank.lee_at_vta.org>
Viktor.holmberg_at_abnamro.co.uk
alan_at_nabeth.cxo.dec.com
jrobens_at_davidjones.com.au


There was no magic answer unfortunately.

The simple answer is to mirror up the volumes before adding them to ASE.
I had actually considered this, but..., our change management people
dont like ANYTHING much being done on a LIVE server unless it is being
done in a scheduled maintenance slot (almost always at weekends). I was
seeking a FAST way of doing it in a maintenance slot.


(I am not particularly an HP fan, but the HP-LVM mirroring does seem to
be better than LSM in this respect...

If you LVCREATE a volume to its full (big) size, then LVEXTEND it by
adding (a second) mirror, LVDISPLAY shows the volume mirrors synching
up. This takes - well as long as it takes- just as LSM does.

However, if you LVCREATE it small (say 4mb), then LVEXTEND it by adding
(a second) mirror, LVDISPLAY shows the volume mirrors synching up, but
they sync up (obviously) very quickly - just as LSM does. However, if
you then LVEXTEND the (small mirrored) volume to its required (big)
size, the extension is almost instantaneous. LVDISPLAY shows both
mirrors are synced up and there is no mirror activity so far as I can
see.

I can only guess that there is some inteligence here which says 'ah -
I'm extending both mirrors simultaneously by the same amount so I dont
need to do any shadow copies'.)


The original posting is at the end of this document


Regards - Tony

+-----------------------------------------------------------------+
| TONY MILLER - Systems Projects - VODAFONE LTD, Derby House, |
| Newbury Business Park, Newbury, Berkshire. |
+-------------+---------------------------------------------------+
| Phone | 01635-507687(local) |
| Work email | ANTHONY.MILLER_at_VF.VODAFONE.CO.UK |
| X.400 | G=ANTHONY; S=MILLER; C=GB; A=GOLD 400; P=VODAFONE |
| FAX | 01635-506709(local) |
+-------------+---------------------------------------------------+

Disclaimer: Opinions expressed in this mail are my own and do not
reflect the company view unless explicitly stated. The information
is provided on an 'as is' basis and no responsibility is accepted for
any system damage howsoever caused.


==================================================================
===Hank Lee ======================================================
Please post a summary on this. I've encountered the same problem
creating
LSM mirrors.
Thanks.

--
Hank Lee <hank.lee_at_vta.org>
UNIX Systems Administrator
SAP Basis Administrator
Santa Clara Valley Transportation Authority
==================================================================
===Viktor.holmberg_at_abnamro.co.uk==================================
No. You haven't missed anything. 
LSM will perform a physical copy from disk1 to disk2 when it mirrors or
resync the mirrors. I haven't tried to grow a volume.
The only option you have is to create a number of smaller mirrored
volumes then stripe them. Of course this defeats the HSZ70 striping.
If I was in your shoes I would create the new volume with the system up,
wait until its finished, then add it to ASE. I think you are being
optimistic with 6 hours. Your main bottleneck is the FWD scsi bus and
the KZPSA controller. It will be even slower if the system is loaded.
If you are using ASE 1.5 you don't need to shutdown the service to
update the LSM configuration. 
Let me know if I can offer any other help.
Viktor
==================================================================
===alan_at_nabeth.cxo.dec.com========================================
If the HP mirroring is so far for a large amount of data,
 it seems clear that they aren't copying the data before
 they claim the operation is complete.  There are techniques
 to keep track of data has and hasn't been mirrored and let
 the startup go very quickly and the actual copying take
 place in background.  But, you have realize that your
 data isn't protected until it has all been copied.  This
 method is useful for a new file system, because most of
 the data is considered free space and doesn't need to
 be mirrored.
 Copying 56 GB of data in 6 hours is around 2.6 MB/sec
 which isn't bad, but isn't great.  There might be
 parameters that control the copy size either in the
 command to make the mirror or in some other command.
 With there being so many LSM command and so many options,
 going through the manual pages and guide to using is
 about the only way to find these things.
==================================================================
===Alan Davis===============================================
  Time to create mirrors depends on several factors.
Size of the volume to be mirrored
Number of plexes being added
Speed of disks
Layout of volume's subdisks (contiguous vs. subdisks
 scattered across disks)
Layout of volumes plexes across disks/controllers
Speed of the buses involved
Amount of disk activity on the volume during the mirror creation
Six hours doesn't sound unreasonable for a 54gb populated
volume to mirror. 
Note that you can continue to access the volume if you are
doing the mirror operation with the service online, but it
will slow access and completion time a certain amount.
You are correct that a mirror operation or resynch will be
interrupted when a service is failed over, stopped or started.
The first thing that happens when you start or failover a service
is the stop scripts are run to ensure the start script begins at
a known configuration.
==================================================================
===jrobens_at_davidjones.com.au======================================
We have 64GB mirrors, and have to confirm your results that it can take
up
to 3hours for us to create a single mirror depending on how busy the
system
is (about 1.5days to do all of them - making breaking the mirrors
inappropriate for backups). Apparently a block copy of the entire disk
has
to be performed when mirrors are created.
We are using clones for backups instead.
We have not seen the problem you describe with TruCluster (and we do
have
one). It might just be due to luck, as we don't rebuild the mirrors
regularly, and from memory, we have concentrated on starting the cluster
after a failure, rather than trying to re-create the mirrors.
==================================================================
===Original posting===============================================
We are about to schedule some down time on a production ASE cluster to
add some additional mirrored LSM volumes to an existing service.  The
volumes are approx. 54gb (HSZ70 stripe sets).  The mirror sets are under
LSM control and mirrored across controllers.
I have done a few tests with creating and mirroring the volumes (on
200000 block volumes) just to get some timings I can use to extrapolate
to the 54gb volumes.  I get some interesting (disappointing) results. 
Let me explain:
#volassist -g tony -U fsgen make testvol 200000 mirror=yes rz14 rz30
Average time=44 seconds.
#volassist -g tony -U fsgen make testvol 200000 mirror=no rz14
#volassist -g tony mirror testvol rz30
Average time 39 seconds - 88% quicker than creating the volume with two
mirrors.
#volassist -g tony -U fsgen make testvol 200 mirror=no rz14
#volassist mirror testvol rz30
#volassist -g tony -f growto testvol 200000
Average time 43 seconds - virtually identical to creating the volume
with two mirrors.
Using the second example, this extrapolates roughly to a little under 6
hours to mirror up!  A bit longer than I had hoped for.
I was disappointed that example 3 did not mirror up much faster.  Under
HP-UX/LVM, creating a volume small, then adding a mirror (with the
volume small) then extending the mirrored logical volume to its required
size is virtually instantaneous.
Does anybody have any ideas how I can speed up the mirror process (apart
from doing it the day before!).  Am I missing something here?
The volumes cant be in the process of mirroring (through unfortunate
experience) when they are added to the ASE service.  The service restart
times out as the disk group fails to export due to uncompleted LSM
activity.
You either have to add the volumes (no mirrors) to the service and add
the disks which will contain the mirror plexes to the service.  Allow
the service to restart then add the mirrors (what we ended up doing). 
Or, mirror up the volumes and wait for the mirroring to complete before
adding them to the service.
Received on Fri Feb 05 1999 - 11:08:25 NZDT

This archive was generated by hypermail 2.4.0 : Wed Nov 08 2023 - 11:53:38 NZDT