Ignoring a possible HA cluster or Oracle RAC cluster for failover, you have several things to consider, at least:-
- Operating system recovery
- Application/Data file recovery
- Database recovery
My basic opinions are listed below.
Operating system recoveryAIX is great because of the built in mksysb command, the authors of which, I'm informed, went on to write the Storix product which is why that is so good and is now available for all sorts of other platforms doing an excellent job. If your server has been laid out sensibly, you will have just the basic minimum to run the OS in the root volume group, rootvg and perhaps just other recovery software. Keep therootvg small or at least empty, because mksysb will try to write the whole thing out to tape/disk/DVD when you call it and it must fit onto one item of removable media, whatever that is. Recovery is from the firmware prompts or perhaps your contracted DR site (if you have one) will do that for you. It will rebuild an OS that matches your current one, including network addresses, so be careful if you try to recover to a live environment.
Application/Data file recoverThe is where your other filesystems would be saved/recovered. Using tar is probably adequate for many, but does not give you all the catalog and history information that a good tool will. There are many, and in my time I've used CA-Arcserve/CA-Brightstore; Netbackup; Legato; TSM; CommVault and others. It is a personal preference on how you get on with them and how easy you find it to recover the catalog so you can actually recover your data. If the server with the catalog is lost, can you retrieve it from the latest tape? How do you know which tape? (Paper listings of tape write dates in the off-site bag are very useful here!)
This should be used to bring back all your filesystem data except databases and should also backup the contents of rootvg, but be careful about restoring over with those bits. Usually you have to have some sort of build process to recreate your filesystems ready for your file recovery to take place. You may need to script up create decks are regenerate them frequently. A good place to store them is in the rootvg so that an OS recovery will give you something to work with. Of course, AIX is actually better than that, so you can backup a volume group structure withy no data and put the (pretty small) file within rootvg so your rebuild is more a case of recovering an empty volume group definition (and therefore all the logical volumes and filesystems) by giving it a list of disk physical volumes to rebuild it on. I think that it even mounts them for you. It is good practice to ensure that filesystems from different volume groups are not dependant on each other. If they are, then the recovery needs to be correctly sequenced else things will go awry.
You may also have to consider software licence keys. If there is a licencing process to follow, you may be as well to store the keys in files in the rootvg too. I know I'm yet again breaking my rule of keeping rootvg minimal, but these could be critical to a recovery and are therefore worth it.
Database recoveryInevitably this should be with the tools supplied with the database. Oracle has rman, Informix has it's own, DB2 has another one etc. and you would be foolish to try to create your own. It's all down to catalogs really. You need to be sure that you will have the latest information for your database recovery to work from, else you will recovery to the wrong point in time and have to start again. I've learnt this the hard way
All that said, backups to disk, if you have the space, can be very good. They are generally quicker than writing to external/removable media, however if your server is lost it can be very difficult, if not impossible, to recreate them so you have to decide what is a 'nice to have' and what is practical in against the risks of disk/server loss.
Why not explore having both? That way you have a potentially quicker recovery from an error and protections from a loss of hardware.
It's become rather a brain dump, but I hope that there is something usul in here to help.
Kind regards,
Robin