topic Re: Copy 50GB data in Operating System - HP-UX

Copy 50GB data

hp_user_1 — Sun, 18 Jan 2009 23:09:54 GMT

Hello,

I need to copy 50GB of data from one file system to another on the same 7410 server running 11iv1. Both file systems are in DMX3, but in different LVM volume groups. I tried cp, but after copying ~2gb it became extremely slow (~3MB/min). I know cpio, rsync, tar, etc. are available, but need advice.

We also tried backup and restore with NetWorker and it took almost 15 hours, which is unacceptable to the customer.

Points will be awarded.

Thanks

Re: Copy 50GB data

hp_user_1 — Sun, 18 Jan 2009 23:21:58 GMT

Hi,

Like to add some more info.

I tried with "cp -rp source destination" and was able to copy 2gb in about 2 hours. There were two instances where cp copied ~900MB in 2 minutes, otherwise the speed range was between 2MB and 10MB most of the time.

Thanks

Re: Copy 50GB data

Venkatesh BL — Mon, 19 Jan 2009 02:46:33 GMT

Did you try 'dd'? It may be faster. Try with a sample file first.

Re: Copy 50GB data

Mark McDonald_2 — Mon, 19 Jan 2009 03:11:09 GMT

Is the 50Gb all of the data? If so, why not mirror it on the frame? Make the 2nd file system a BCV of the first?

Re: Copy 50GB data

hp_user_1 — Mon, 19 Jan 2009 12:50:42 GMT

The BCV copy is not an option. Also this is one time requirement.

Only option available is to copy it at the OS level, as NetWorker is too slow. cp gets slower overtime, don't know why. I changed certain parameters on the destination fs with vxtunefs, but not luck.

Thanks

Re: Copy 50GB data

TTr — Mon, 19 Jan 2009 13:08:24 GMT

> I tried cp, but after copying ~2gb it became extremely slow (~3MB/min)
> 2gb in about 2 hours...
> cp copied ~900MB in 2 minutes

This fluctuation in your copying process may be the result a several scenarios
1. You may have soft or hard read/write errors that only manifest in certain times or disk areas. Check the server logs for any errors. Look for SCSI errors (lbolt) or read errors. Check your source copy area for read errors and your target copy area for write errors using dd commands.
2. Your LVM volumes/filessytems have hot i/o spots. This depends on the LV and PV layout on your server as well as the LUN setup in the disk array.
3. Your server may be running our of resources after the copying starts. Check you memory usage, disk and swap utilization etc. Use glance if you have it or top, sar, swapinfo etc.

Re: Copy 50GB data

sujit kumar singh — Mon, 19 Jan 2009 13:08:32 GMT

hi

can u give a try to fbackup / frecover using s config file!!!!!

regards
sujit

Re: Copy 50GB data

Steven E. Protter — Mon, 19 Jan 2009 13:28:51 GMT

Shalom,

Options:

1) scp -rp via the NIC card. That should not slow down.
2) Connect the arrays and use array utilities. (Not an option on the 7410 I think)
3) fbackup, carry the tape, frecover

SEP

Re: Copy 50GB data

hp_user_1 — Mon, 19 Jan 2009 14:25:33 GMT

The source and destination file systems are both on the same server...

Re: Copy 50GB data

Peter Nikitka — Mon, 19 Jan 2009 15:04:09 GMT

Hi,

I think there is no much of tuning possible, when the two LVs do already exist.
Is the data you want to copy the whole content of the filesystem?

Always possible:
cd Parentdir
find dir | cpio -pud /target/directory

Whole filesystem:
cd /target/directory
vxdump 0f - /file/system | vxrestore xf -

mfG Peter

Re: Copy 50GB data

Olivier Masse — Mon, 19 Jan 2009 16:41:04 GMT

What kind of data are you copying? If you're talking about thousands of files and only a small percentage of them actually change from one time to another, use rsync and you'll save lots of time. Just rsync once, stop your application, then rsync again to copy your delta, and restart your app on the new filesystem.

Re: Copy 50GB data

hp_user_1 — Mon, 19 Jan 2009 16:51:21 GMT

Thousands of files, and maybe 20% of them change frequently....

Re: Copy 50GB data

hp_user_1 — Mon, 19 Jan 2009 16:52:07 GMT

Need to copy one directory structure only, and not the whole file system.`

Re: Copy 50GB data

sujit kumar singh — Mon, 19 Jan 2009 19:27:58 GMT

hello,

as this is local copy even cp and cpio, tar shall take a lot of time to backup or copy or restore.

dd or vdump as i feel u cannot use in this case as u are not to copy an entire LV or a Mounted LV.

dd would have been surely faster with an optimal and big Block Size specified.

i think u can go with the help of a fbackup and frecover with a config file. That shall help u open multiple readers processes in the memory and also help u make the use of fastness of fbackup by specifying a big block size hopefully.

cant tell of other backup softwaRE but Data Protector would have been really handy in this case.

however, using fbackup with a config file, u can achieve a better performance in backup and restoration, u can try like this.

u can have a look at the thread given below and read the post of JRF

http://forums11.itrc.hp.com/service/forums/questionanswer.do?admit=109447626+1232392329851+28353475&threadId=1296457\

and see this thread

http://forums11.itrc.hp.com/service/forums/questionanswer.do?threadId=815533

see the last as Bob Brown says:

Feb 17, 2005 11:56:37 GMT N/A: Question Author

--------------------------------------------------------------------------------
Well...I experimented and got MUCH better results. I'm up to >40GB/hour right now.

I set blocksperrecord to 8192,
records to 64,
checkpointfreq to 4096

Those made the biggest difference.

-Bob

u can achieve this by creating a config file for fbackup like this

#vi /tmp/fback.config

blocksperrecord 4096 ---- or 8192 ..... bob got a good result almost 40 GB /hr on Ultrium tape
records 64
checkpointfreq 4096
readerprocesses 6
maxretries 5
retrylimit 5000000
maxvoluses 200
filesperfsm 2000

save and exit.

then use this config file to take a backup.

#fbackup -0vf /dev/rmt/0m -c /tmp/fback.config -i /

here 0 mens full backup.
-f /dev/rmt/om is he device file of the Tape Drive.
-c /tmp/fnack.config is the config file.
-i / is the directory u wish to take a backup.

and while restoring the same,

rewind the tape:

#mt -t /dev/rmt/0m rewind
go to the directory u wish to recover

#cd /test

say

#frecover -rvX -f /dev/rmt/0m -c /tmp/fback.config

or #frecover -xvX -f /dev/rmt/0m -c /tmp/fback.config -i /

-r is for recover all from the tape
-x is for selective recover of files.
-v verbose
-X is for recover to current directory instead of recovering to the original location.

thanks JRF abd Bob for the informations...

Regards
Sujit

Re: Copy 50GB data

sujit kumar singh — Mon, 19 Jan 2009 20:12:01 GMT

Hi

also make sure that ur system running 11.11 has the latest patches for fbackup, that will be an added advantage ... u can go to the itrc .. patches ... select .... 11.11 ...search by keywords "fbackup", downlaod and install that later on also ...

that helps fbackup exceedingly enhance the functionality with the New Ultrium Drives and Media.

however u can do that later on also ..

Sujit

Re: Copy 50GB data

Peter Nikitka — Mon, 19 Jan 2009 20:32:46 GMT

Hi,

you described the task as a one time job.
Using temporary storage is - in my feeling - not the adequate technique for copying files from one file system to another (on the same server!):
Using temporary storage involves an additional read and write operation. This overhead cannot be regained, IMHO.

I still think,
find ... | cpio ...
is hardly to beat here.

mfG Peter

Re: Copy 50GB data

sujit kumar singh — Mon, 19 Jan 2009 20:46:58 GMT

Hi

Thats being true, that the most suitable command for such operations is definitely the combination of find and cpio

that is '

#find ..... | cpio ....

and that is really good for the fileystem copy as we cannot use dd, and vdump here ...and also that cp is slower.

yes the #find .... | cpio ....

that is also a really recommendable solution here as surely that will not require any third storage and that this is more flexible too.

better u can as suggested also give a try to cpio also.

I fully agree and that is more flexible too.

Regards
Sujit

Re: Copy 50GB data

hp_user_1 — Tue, 20 Jan 2009 00:10:52 GMT

I tried "cd source; find * | nice --2 cpio -pudlvm /destination" and it worked much faster than cp, but still I did not get the speed I was looking for. I got about 8GB / hour .... I need to copy the 50GB in less than 3 hours in prod window ....

With fbackup/frecover, can I fbackup files and pipe the output to frecover to restore in destination without using tape?

Re: Copy 50GB data

sujit kumar singh — Tue, 20 Jan 2009 00:48:47 GMT

Hi

http://forums13.itrc.hp.com/service/forums/questionanswer.do?admit=109447627+1232412006748+28353475&threadId=1047319

see the post of marvik ::: U can use also dd with an optimal blocksize::

just FYI incase u want to copy large FS in the same server use:
umount /fs1 (X series)
umount /fs2 ( Y series)

dd if=/dev/vgxx/rlvolx of=/dev/vgyy/rlvoly bs=1024k

see another in the same post by marvik

# cd srcdir && fbackup -i . -f - | ( cd dstdir && frecover -Xsrf - )

Now its ur call,u have many options :)

also see the paost from JRF

Regards
sujit

Re: Copy 50GB data

Steven Schweda — Tue, 20 Jan 2009 00:52:22 GMT

> I tried "cd source; find * | [...]

"find ." might be smarter than "find *".

I'd probably use a "tar" (or "pax") pipeline,
but I haven't run any speed tests, so I
don't know if it would be faster than cpio or
anything else.

A Forum search for "tar pipeline" should find
many examples in previous discussions.