backups and cloning strategies with arrays

[Francisco Quesada]

Hi all.

I've been searching in several groups for answers to my doubts in managing arrays and so on. These are my conclusions and my doubts. I would like to receive confirmations and corrections. Let's go:

We have a Compaq ML370G2 with a 5302 smartarray controller, with 2 36Gb disks in RAID 1. The server
uses W2000 Server and it's mainly a database server with sql server and Informix. It's not a 24x7 server so i can shutdown it sometimes.

Backups to tape are made daily at night. This point is not a problem, with Veritas or even NTBackup i
can make copies of data coming mainly from both Informix and Sql. These point is clear for me. But i
need a method against failures of software upgrades or software installation. And also for a total disaster.
I know three methods for this, but maybe each method process is not clear for me. These are:

(note:i know disks are hot-plug. I shutdown the server because i like more).

1) Changing disks method: I have a third disk identical to the others. I shutdown the server. I extract disk 1 on slot 1. I connect the third disk on slot 1 and start the server. The rebuild is made. I have in my hand a disk with an exact copy and being part of the array. Now i make some software upgrade on the server and something goes wrong, and the system is corrupt. To solve the situation: I shutdown the server. I extract both disks on it. I insert the disk, on slot 1, with an exact copy extracted before.
Now if i start the server i suposse that it could start in degraded mode and everything is ok. Or this is not true?
Then i shutdown the server. And now the question: if i introduce on slot 0 the disk that was on that place and that it has a corrupt software image .... wich rebuild is made? The correct rebuild would be disk on slot
1 onto disk on slot 0. And why? because is some kind of "master" if it is older connected to the array than the new one?

2) Using some cloning software, for example, Ghost 7.5 : Symantec warns that ghost not support raid disks. But on several groups i've read examples of cloning with raid, but not too clearly. Some people say that if raid is trasparent to the Operating System you can clone the system. I suposse this is not true because if it was true i don't understand the Symantec warning. I think that to use this method it is necessary to break the mirror, clone one disk sector-by-sector and reconstruct the mirror again. Only in this manner i can have a correct image useful in case of restoring the system. The problem is that i think i would prefer not to break the mirror ... what do yo think?

3) Using some backup software with support for disaster recovery, for example Veritas. I have an AIT tape for daily backups. It support "intelligent disaster recovery" (IDR) , and Veritas support it as well as an option. So the theory says that i can make a complete copy of the system on a tape using Veritas and IDR. Then, if a disaster ocurr i can restore the system inserting the IDR tape and starting the server from it, and WITHOUT USING ADITIONAL SOFTWARE!
So if i make one of this IDR backups with Veritas on a tape. Then i install software on the server and something goes wrong and the system is corrupt. Then i want to restore the system using the IDR tape made before. I start the server
with the IDR tape inserted and WITHOUT USING ADITIONAL SOFTWARE AND WITHOUT CONFIGURING NOTHING AT ALL the system is restored correctly.
I think this don't work but i am not sure. I don't know if an IDR tape has the same amount of data and the same
functionality as a cloned image from ghost. Anyone knows Veritas and IDR well enough to explain clearly?


If someone wants to comment some or all the methods, or knows some place in Internet with deeply information
i would apreciate.

Thanks.

[e4services]

We like option 1.

Actually it is a process we use all the time and a nice thing about a mirror, you get a functioning standalone drive. As for the re-introduction of the mirror, we have no answer. We delete the partition on the introduced drive prior to inserting it as a practice, but not by knowledge. We have never tried to see if there is a presidence to the controller. I would assume if done "hot" it would know which is the newly added, but if cold, now were down to firmware. We always assumed the developer assumed a New drive, no data, as a replacement, so that is what we go one. It is not much of an effore to delete a partition before.
As for Partition Image, again assumptions, Raid 1 mirror should be straight forward and present no problem to the software, but RAID 5, you would think, gets a bit more complex. I think this is why PowerQuest always made PartitionMagic a client device, not a server device and wrote a special product for server.