Moving Large Mount Points of Data

[Alice Round]

I've been testing different ways of moving data and the following results are what I've came up with: I'm looking for the fastest
way to move large mount points.

cd /baant_new
find . | cpio -pudmvl /var/crash/test

22 minutes this moved 3 GB

cd /baan_new
cp -Rp * /var/crash/test

22 minutes this moved 2 GB

I used Omniback 2.55 to backup /baan_new which was 8.5 GB to tape this took 45 minutes

Then I restore /baan_new to /var/crash/test using Omniback this took 1 hour and 12 minutes.

P.S. This thread has been moved from Storage > General to HP-UX > sysadmin. - Hp Forum moderator

[Rick Garland]

Not sure about the speed but I would also consider dump/vxdump to be an option as well. Will keep the perms and the access times.

[Alice Round]

dump and vxdump only puts the filesystem
to tape. I'm really wanting to move large
mount points to a larger mount point.

[Brian M. Fisher]

I know this might sound like a silly question but have you tried the move command (mv)?

Brian
<*(((>< er

[Alice Round]

We aren't using the mv command because we want the data left where it's located for recovery.

[Kofi ARTHIABAH]

did you try using tar or pax? eg.

cd fromdir ; tar cf - . | ( cd todir ; tar xf -i )

[Rick Garland]

dump/vxdump will put anywhere you want, including another disk or another system

[Antoanetta Naghiu]

dd command could be a option... I never tried it disk-disk but should be fast.

[Brian M. Fisher]

dd is a very good option if you want to quickly copy data from one place to another. The output file needs to be the same or of greater size than the input file. The block size option should be used due to the default being 512 bytes. An example is as follows:
dd if=/dev/rdsk/c0t15d0 of=/dev/rdsk/c0t6d0 bs=1024k

I'm sorry about my first response, I was under the assumption you wanted to "move" the data.

Brian
<*(((>< er

[Dave Wherry]

Do you have Mirror/UX? A possibility is to mirror the data and then do a lvsplit. Then mount up the split copy of the mirror to your new mount point. Then if you have On-Line JFS you could lvextend your volumes.
There is a document on the ITRC - ID X1401978- that discusses doing this for booting off of a split mirror.
I did this several times while testing a UX 11.0 upgrade. It might work for you and it would be 95% on-line.

[Anthony deRito]

Don't forget, dd is a real dumb copy. It copies every single block in the device whether there is data or not.

What I would do is stick with your cpio. cpio is fast and works very well. You may want to change it to:

#cd $INDIR
#find . -xdev -depth -print | cpio -pxdmu $OUTDIR

Then look at performance issues around the data transfers. Are you doing sunchronous file transfers or asynchronous file transfers? Look at your fs_async kernel parameter. You may want to change it for async FS I/O. This will dramatically speed up transfers. (beware the caveate as per SAM help ...)

Also, check out sar -d for your buffer cache hit ratios. Look at the %rcache and %wcache values to see if your getting good use out you buffer cache. If not, check out tuning your buffer cache.

Also check out your sar -d values. If your using a disk array, think about adding some add-on cache to the service processors. This will boost up the service time of the disks and possibly the transfer rate as shown by blks/sec.

Hope this helps.

Tony