fast backup

[KINGSLEY_1]

Hi!

We are running a banking application which database is oracle. Data protector takes almost 2 hours to finish backing up before we can start close of business.

Is there a way where we can have a copy of the application and database so that the close of day would not wait for the backup to finish.

Again, if there can be a copy can the copy replace the original in case the original corrupts.

Thank you.

kingsley

[James R. Ferguson]

Hi:

Snapshots (of VxFS) filesystems are designed to allow you to do "quiet" backups while continuing to read and write.

Regards!

...JRF...

[KINGSLEY_1]

Hi Ferguson!

Our file system is (LVM). In fact, i read somewhere that snapshot becomes useless when the source corrupts.

At the moment what we do are these.

1. Before Close of business the database has to be shutdown.

2. We would spend 2 hours to take the backup

3. After the backup, the database is brought up again.

4. Close of business is run

5. After close of business another backup is run.

Now we want to know if we can copy these files to another place using shorter time so that we can give way for close of business to start whiles we start the backup from this new site.

Thanks

Kingsley

[Steven Schweda · Accepted solution]

> [..] Data protector takes almost 2 hours to
> finish backing up [...]

Backing up how much stuff to what?

If you're copying to a tape, then it might be
faster to copy to a disk (and then make a
tape from that, if you want a tape).

I don't do it, but I'd expect that using a
disk mirror would let you break the mirror,
and then you could play with one piece with
no significant delay. (The less activity
before re-forming the mirror set, the faster
the merge operation would be.)

[James R. Ferguson]

Hi (again):

> Our file system is (LVM).

The Logical Volume Manager isn't a filesystem. You may have filesystems within logical volumes and if those filesystems are VxFS ones (almost guaranteed these days) then you could use the JFS (VxFS) snapshot.

If you would like to learn more, one place is the manual URL listed below. Look at page 694 and beyound, in particular.

http://bizsupport.austin.hp.com/bc/docs/support/SupportManual/c01920025/c01920025.pdf

Regards!

...JRF...

[Emil Velez]

Your backup can be configured to use the disk array snapshot and snapclone functionality and mount a snapshot on another system so as far as your original system is concerned you are finsihed once the snapshots are created.

This can be done for oracle and filesystem backups.

[Michael Steele_2]

What you are describing is how companies like EMC make beaucoup bucks through BCV's. BCV's are not new and are specifically designed for this type of problem. And, the the database is down for only 20 minutes instead of until the backup finishes.

How do BCV's work. A third mirror is taken, sync'd up, split off and then the backup starts on the 2nd mirror while the database starts up on the primary and alternate.

As mention up a similar solution exists at the file system level with vxfs. You copy one entire file system over to another and backup from there.

You can add a 2nd mirror disk via LVM and split it off through your own custom script. LVM is the level below the file system.

On the backup side there will always be bottlenecks. You can add in a high speed tape drive like the Ultrim 960.

And of course, start everything by making sure you are up to day with firmware, OS and diagnostic patches.

[KINGSLEY_1]

Hi Ferguson!

The bad news for me is that we do not have the lincense for backup to disk.

Does the file jukebox functions like the disk?

In terms of speed, can you compare the two?

Thank you.

kingsley

[Steven Schweda]

> The bad news for me is that we do not have
> the lincense for backup to disk.

License for what? Does "cp" require a
license?

> Does the file jukebox functions like the
> disk?

What's "the file jukebox"?

> In terms of speed, can you compare the two?

Compare "the two" what?


> Backing up how much stuff to what?

Still wondering.


You may know what you're doing, but to some
of us non-psychics, it's still a mystery.

[Michael Steele_2]

"...Does the file jukebox functions like the disk?..."

Jukeboxes are optical WORM devices. WORM means write once read many.

Jukeboxes are not akin to hard drives, they are akin to CD devices. So, like any CD device, expect a 20 times decrease in read transfer rates from a jukebox.

To test this, load in a CD and find a large file. Copy the file first to the hard drive and count the time with timex. Then copy the same file from hard drive to hard drive. You'll see that the difference is huge.

[KINGSLEY_1]

Hi Steve!

I mean data protector lincense to backup to disk before moving to tape.

Some members adviced that since back up to disk is not working i could configure file jukebox to serve the same purpose.

I was asking whether backing up to file jukebox can be compared to backing up to disk as far as speed is concerned.


Daily i backup about 300GB of data and it takes about 2hrs to tape. I am therefore look for a way to cut down the time.

Thank you.

kingsley

[Michael Steele_2]

About your only alternative then is to fool around with block sizes: Both on the tape and in the file system. For a raw device 512 bytes is the default block size. For a vxfs file system 1024 bytes is the block size.

Question: Are your filesystems 'largefilesystem'?

For a tape device it depends, for fbackup its 8192 bytes.

Here are some good links to review, but note, don't expect huge performance changes. Not the kind that you'll get when you upgrade to the biggest and fastest tape drive, or change to SAN from scsi, etc.

http://h30499.www3.hp.com/t5/ITRC-Data-Protector-Forum/Data-Protector-6-0-amp-Block-Size/m-p/4026666#M68446

http://h30499.www3.hp.com/t5/ITRC-Data-Protector-Forum/Ideal-block-size-and-segment-size-for-LTO-4-drive/m-p/4650791#M97082

http://h30499.www3.hp.com/t5/ITRC-Data-Protector-Forum/Blcok-Size-In-Data-Protector/m-p/4580596#M94176

http://h30499.www3.hp.com/t5/ITRC-Data-Protector-Forum/Dataprotector-6-0-LTO4-and-1024KB-blocksize/m-p/4289069#M81431

PS - Also pose this thread in the data protector forum - Maybe you'll get better info.

[Steven Schweda]

> [...] i could configure file jukebox [...]

I still don't know what "file jukebox" means
to you.

> Daily i backup about 300GB of data and it
> takes about 2hrs to tape. I am therefore
> look for a way to cut down the time.

Ok. I assume that it would be faster to copy
these data to some other hard disk storage
than it is to copy them to tape. If that's
true, then why not copy the data to some
other hard disk storage (using some simple
method, like, say, "cp" or "tar", or some
fancy method like using a disk mirror), and
then use Data Protector to put these _copied_
data onto tape, while you go back to using
the normal data? You still need to stop
writing to the normal data while you make the
disk-to-disk copy, but after that's been
done, you don't care how long it takes to
make the tape (so long as it's less than a
day).

You also get a fast (on-disk) back-up source
to use while the tape is being made.

[Jose Luis Casado]

Hi,

Maybe it is worthy for you to check the possibility to buy the Data Protector license that will allow you to run oracle online backups.

If this not an option, then you can create a set of scripts that:
1. Put the database on backup mode.
2. Create an VxFS snapshot.
3. Remove the database from backup mode.
4. Backup the snapshot.
5. Backup all the archives created during this process.

That should guarantee that you have a valid backup.

Best regards
jose

[klb]

Sounds like the OP is not running the Oracle instance(s) in ARCHIVELOG mode, so setting backup mode isn't an option.

I would do this as a quick fix:

1) Create a new 300GB filesystem

2) Write a script that copies all Oracle datafiles and ONLINE redo logs from their production locations to the new 300GB filesystem. This script will probably need to do this with multiple threads in parallel like:
cp file1 $NEWFS/backup/file1 &
cp file2 $NEWFS/backup/file2 &
cp file3 $NEWFS/backup/file3 &
....
cp filen $NEWFS/backup/filen &

run as many of these in parallel as your I/O subsystem can handle to gain throughput and just keep chugging through the files until they're all copied.

3) Modify the nightly processing steps as follows:
o shutdown database
o fire up the script and make the copy
o once copy complete...
o start database and begin close of business
o while COB is taking place... continue ...
o start the backup of the copied files to tape ( note this backup will now simply be a file backup rather than an Oracle database backup, so you might also want to get a copy of the other things Data Protector would normally get like the parameter file ).
o Once COB is done, you can then get your final backup either using this new method or using the standard Data Protector backup.

This would effectively reduce the time required to get an offline backup, probably quite a bit depending on what type of disk subsystem you're running.

Hth,

-klb

[klb]

How many tape drives does Data Protector have access to?

If more than one, try to configure DP to use all of them at the same time in parallel. I've used this approach with other backup software ( Alexandria and NV ) with success. Basically you end up sending multiple streams onto tape reducing the total time required to get it all on tape.

Not sure w/DP whether its possible to have a single backup job that writes to multiple devices in parallel?

Hth,

-klb