Solaris

We use SW mirroring (RAID1) using SVM for our SUN servers (which we OEM to end clients with our application on). When I need to make a tape backup for the server (disk), I disable mirroring and perform backup of the entire disk to tape using ufsdump. This process requires an outage and sensitive as there is no room for mistakes. It is relatively complicated to other engineers/clients to do this and support SW mirroring using SVM on regular basis (as experienced in the last few years supporting this). So I decided to got for HW mirroring using raidctl which seems to be more straight forward and reliable. However, I am exploring option to make tape backup strategy for the whole disk with minimum outage and relatively easy steps/procedure. I can think of FLAR to tape... but can anyone suggest something based on experience that is reliable and easier to do?

I would appreciate it if you can refer me to way or some documents about that... maybe Sun doc

I have a preference not to use raidctl to be honest - it removed any flexibility for you to temporarily use that disk in the event that you need to.
We regularly make FLAR archives of system mirrored under SVM without having to split the mirrors.
Your other option, although can be costly, is to invest in a proper backup strategy and use something like NBU or Solstice Backup (Legato Networker).

As to using HW mirrored disks - the OS does not see them as individual disks so just treat them as a single disk for the purposes of backups.

Since you are on Solaris, use fssnap and ufsdump on the snapshot. That would be partition by partitiion, not whole disk. You could set up a script that handles it, put it in an adm directory (I use /usr/local/adm/backup), and then just run that script. Set up permissions so only your engineering logins have access to it.

I've used svm in the past, but with my new T5220's, I'm using raidctl for the mirroring of the root drives. My script for doing backups works fine. Amanda also works well using the same tools. However, it sounded like your situation is more of a one off thing when you are distributing systems.

If you are running realtime processes (like ntpd), you need to stop those before doing fssnap and then start them up again. That can be incorporated in the script as well.

Another reason that made change to raidctl is using and implementing LDoms. I found it a bit complicated maintaining the server with SVM and LDoms. I had to come up with different strategies to make it work (since Sun only provides very simple and basic examples). So I moved to raidctl to get rid of one of the complication factors (i.e. SVM). Costly options are not an option now under the current financial circumstances. FLAR is also an option if it would allow me to make a full backup of the disk. I found a lot of material about FLAR, but can you point me to something that talks about full disk image?

choogendyk, I am also using T5220 (and T2000). Can you share your script so I do not have to re-invent the wheel? With regards to our business model, we install our application on Sun servers and sell them to our clients (OEM the HW); so we are responsible for the support. Since we partition the disks according to the application needs and manage mirroring, we have to come up with the backup strategy as well. In your case, you bring root partition down to make the backup, right? If that is the case, this means service outage in my case.

Actually, no, I don't bring down the system to make the backup. That's the point of the fssnap. It reduces your exposure for changes to just the time it takes to make the snapshot rather than the time it takes to do the ufsdump.

I'll see if I can get a clean copy of my script, a bit of documentation, and an example config file up on the web this weekend, and then drop some links here. The script is several pages long, and the latest changes I made were to accommodate Solaris 10, using smf rather than init.d to stop and start ntpd.

Thanks, I am not very familiar with fssnap; I am searching for material about it. Thanks in advance for providing the script.