Backup Options

[Marcos Trinidad]

I am currently using the LoneTar utility on my k370 with HP-UX 11.0 to perfrom nightly backups. I am backing up vg00 and vg01 (data and application) on to a DLT tape. vg01 contains a flat file database (DBC). An iginite tape is made once a month. Currently I have to log off all users to perform the backup. I want to be able to perform a full online backup without having system downtime and it still be consistent.

1. What is the best way to accomplish this?

2. Will purchasing Online JFS or Omniback II solve my problems?

Thanks in advance.


-Marcos

[A. Clay Stephenson · Accepted solution]

The particular backup package you choose has very little to do with your problem. Whether or not an online backup can be made is almost solely determined by the applications you are running.


As a complicated example, I will use our ERP system. It consists of apllication servers which contain meta-data that describes the actual data in an Oracle database. In this case, even a hot Oracle backup is useless UNLESS the matching meta-data is kept in sync with it.

More than choosing a backup product, you need a backup plan and that involves knowing your particular set of applications.

One thing that OnlineJFS does buy you is the ability to do snapshot mounts. With this, you could shutdown your application(s), make a snapshot mount, and restart the applications. Typical application downtimes are a few minutes (typically under 5 minutes). You then backup the snapshot mounts. You could do something very similar with Mirror/UX by spliting the mirrors.

OmniBack is a pretty good choice because you have the ability to easily add pre and post exec commands to automate all these steps. When used in conjunction with a tape library, you can have fully automated backups.

[Michael Tully]

Your backup software a) has to have to capability to be able to backup in 'hot' mode and of course your catabase has to have the capability of doing so.
Both On-Line JFS and omniback seperate in their designs could both be used, On-Line JFS can have snapshot mounts, you could use Mirror/UX to do something similar by splitting the mirrors.
Omniback could also be used to this task, by splitting the mirror as a pre-exec job and re-syncing as a post-exec job.

[Bill Hassell]

The question can only be answered by the software that you run. Typical database programs require several files and information stored in one file relates or points directly to data in another file. The problem is that you can't guarentee that the files will be written to the tape instantly--it takes time to read all the data, and during that time, data which was already written to tape gets changed along with other data that is about ready to be saved...and you have a mess on the tape.

Modern database programs have a backup or read-only mode. You signal the database master program that a backup is starting and all volatile data in memory is flushed to the files, then no further updates are allowed to the database until the backup is finished. The other feature of modern databases is that in backup mode, changes are still allowed but the details are written to a special file. Then, once the backup is completed, the change file is played back to update the database.

Another option for some database programs is the ability to use mirroring as an instantaneous backup snapshot. More on this later...

Answering the 2nd question first, no, JFS will not help. JFS simply keeps the directories and individual files intact. JFS knows nothing about relationships between data records inside the files. The snapshot feature can be of some use, but the size of the snapshot file is a consideration. OmniBack has a number of plugins for popular database programs that can accomplish this task but if your database is home-grown, OmniBack won't provide any safety for multi-user backups.

One solution that is quite effective is mirroring. If this is important data, it should already be mirrored in realtime, either with a disk array or with Mirror/UX software. However, many disk arrays (and Mirror/UX) provide an option for triple mirroring. The 3rd set of disks provide a way to take the database offline for a few seconds, split off a complete copy, and then go back online. The split 3rd mirror is then fsck'ed and mounted to temporary moutpoints where the data may be backed up at your leisure, often during the middle of the day (no night crew needed to change tapes.

3x disk space? Yes, and it is incredibly cheaper than the alternatives. A backup solution is an insurance policy and to choose the right policy, you start with a value for the data--is it worth one million dollars, or 50 million? Ignore the cost of the computers and disks; that data, should it be lost, could cost your company millions to replace..and that's the figure you use for insurance. Now, a 1% pricetag for triple mirroring looks pretty cheap as insurance goes.

[George Petrides_1]

I have used mirror-splitting on more than a dozen systems and worked great (both Mirror-UX and Business Copy). If the merging time is not that big, then you can also keep the offline copy until say 5PM and if you lose all your data between the split time (e.g. 2AM) and 5PM, you can switch back to the 2AM data within a minute or two, by mounting the split mirrors and doing reverse merging. Now regarding Online JFS snapshot, although cheaper just keep in mind that the way it works is setup pointers to the data blocks. If a block is about to be modified by the application, then Online-JFS replaces the pointer in the snapshot with the actual data block from the live application data. Therefore, if you will have intense operations (batch jobs, etc) going on during the backup, that they will significantly slow down since every write requires copy of the existing data to the snapshot before the new data is written to the disks. Mirror-UX is simple and relatively fast.
Just my $0.02.
George