Concurrent Restores/Backups

[David Goza]

My full system backup takes about 2.5 hours to complete and a restore takes about 6.75 hours. I have a K450 Box with an external Jamaica Enclosure and 4 Dat 24 DDS3 Tapes Drives.We are on HP UX 11 - 32bit. I have thought of dividing up my main directories and doing 4 concurrent backup or restore operations to all 4 drives at the same time using an "&" at the end of each backup or restore command line. The big question - if it can be done, will it save time?? If this won't work, what will (other than Omniback)? Thanks

[John Palmer]

Hi David,

As the main time limiting factor with backups is the speed at which data is written to (or read from) tape then I would expect your backup time to reduce to about 40 minutes if you run four concurrently.

Your main concern would be how to split up the backups into 4 equal chunks, this would depend on what utility you are using.

I have used this approach numerous times in the past. The only downside that I can think of is that you need four times as many tapes!

Regards,
John

[A. Clay Stephenson · Accepted solution]

Hi David:

There is no way to answer this question because it is dependent upon your data. If your backups are tape device bound then yes it will help but if you are bound by disk I/O then multiple processes will almost certainly be slower.
In your case, using DDS3 I suspect that you are tape drive bound but only doing a trial will tell for sure.

If you are going to drop multiple processes into background, I suggest that you do it something like this:

bkup1 &
bkup2 &
bkup3 &
bkup4 &
wait

The wait will cause the parent process to wait fore all the background to finish before proceeding and thus you have some means of knowing the process finished.

While this will not speed up your restores, let me give you a plan B than will make your backups almost instantaneous.
Why not use vxfs snapshot mounts. This process takes only seconds and then you have nice, stable data that you can backup at leisure. Your regular processing can continue while you backup. When the backup is done, you unmount the snapshots.

[Steve Steel]

Hi


Since fbackup works well and has a graph file to define each backup use that.

See man fbackup frecover.


You can also schedule the backups via sam.


Steve Steel

[David Goza]

Is there documentation on the vxfs snapshot mounts? By the way, I am using the fbackup command to do my backups (in a cron job).

[Sanjay_6]

Hi David,

Try this link from the docs site,

http://docs.hp.com/cgi-bin/fsearch/framedisplay?top=/hpux/onlinedocs/B2355-90742/B2355-90742_top.html&con=/hpux/onlinedocs/B2355-90742/00/00/51-con.html&toc=/hpux/onlinedocs/B2355-90742/00/00/51-toc.html&searchterms=snapshot&queryid=20020207-080638

Look for "Backing Up a JFS Snapshot File System"

Hope this helps.

Regds

[Sanjay_6]

Hi David,

Here is a lin from itrc, which deals with snapshot backup,

http://us-support.external.hp.com/cki/bin/doc.pl/sid=404956410caeee7380/screen=ckiDisplayDocument?docId=200000049791102

Hope this helps.

Regds

[Bill Hassell]

The answer is: it depends. You will never get 4x the throughput if the tape drives are on the same bus. The burst rate (data flowing across the bus memory-to-memory) can be fairly high but eventually, all the handshaking will limit throughput. And don't put any of the tape drives on thew same bus as the disks (same problem, the bus will be too busy to keep the drives busy)

So add at least 2 additional SCSI cards for the 4 drives, 4 SCSI cards would be ideal. Then make sure fbackup is running with a config file that looks like this:

blocksperrecord 256
records 32
checkpointfreq 1024
readerprocesses 6
maxretries 5
retrylimit 5000000
maxvoluses 200
filesperfsm 2000

Now the only problem is making sure all four tape drives are going blinky-blinky (technical term) all the time. That will in large part depend on the speed of the processors (note: plural) and speed of the disks. If all that is done, you should see a full backup in 1/4 of the time.

But your mileage may vary...