Years ago I had the pleasure of getting called out to a client site where they had some problems with their IBM Server. I needed access to the data from their Raid 1 pack running on an IBM ServeRAID controller. For some reason that eludes me there was a problem on that machine. “No Worries” I thought, I’ll just drop it in the box beside it and read off the data. It’ll show up as a normal disk on a normal SCSI controller.
Err – NO
You see whilst Adaptec and others do genuinely mirror their disks, IBM implements some type of Mirrored Striping. I don’t know and haven’t researched all the details, but the short story is that if you want to read the Mirrored disk – you need an IBM RAID Controller. I did speak to an IBM tech and was told it was for performance reasons.
Now using an IBM ServeRAID controller to read the data isn’t the end of the world – just throw it in another box (assuming you have one). At least IBM RAID is compatible across their controller model ranges.
This is where your world will now get “interesting”. You see IBM has this interesting concept where they store a copy of the RAID controller configuration on the disks. If you install RAID Pack 1 from Server1 into Server2 that already has a RAID pack (Raid Pack 2) operating, the RAID Controller (assuming there is only one for that backplane) will see two different configurations. Pack 1 knows only about itself and Pack 2 is the same.
Fair enough you say – we’ll just teach them about each other. Well – you can’t. IBM does not allow “merging” of RAID pack configurations onto the controller. This means that if you try, it’ll delete one of the packs. Fun Fun Fun. Great design effort boys.
I haven’t seen this much brains used since Intel decided to make a RAID controller with no recovery console except for the one installed in the OS – THAT YOU CAN’T BOOT to because that’s why you need the recovery console. (I use these for doorstops)
OK, so you can’t merge, so you can’t easily swap packs between machines for quick recovery. Good marketing move, it does give a reason to keep a spare IBM box in the corner.
It also makes one of my safety tricks useless. I have on more than one occasion seen an administrator delete the wrong pack or disk on a machine with several arrays. I haven’t had it happen, but it’s entertaining to watch. It’s not that hard to do. Often the logical labeling is not the same as the physical labeling, or is inconsistent in some other way. My solution to this (assuming things are offline) is to drop the healthy pack out whilst working on the failed one. It’s very hard to make a mistake and delete something that isn’t plugged in.
BUT – if you try this with an IBM ServeRAID controller, the pack you are working on will likely lose knowledge of the offline pack. When you go to plug it back in – presto – pick one, but not both. You then start all over with the healthy pack online, and waste another two hours. Wonderful during a DR situation.
So – if you are used to other RAID manufacturers kit and get dumped with an IBM system, things are a little different.
I actually like the IBM ServeRAID kit, I just get frustrated by companies that insist on being different and make my life harder in the process.