(Warning: big post follows. My apologies if this should've been posted in the General forum, it doesn't get any traffic!)
I'm planning/considering migrating from MetaSAN to XSAN. The reason as to why can be the subject of a separate discussion, let's just say that I've had some negative experience with MetaSAN and our existing infrastructure is perfect for XSAN.
Uninstalling metaSAN should be pretty darn easy, but "Apple's way" and doing things by the book assumes you prepare the infrastructure, then the clients, then the secondary MDC, and finally the primary MDC with the XSAN wizard and volume configuration. I'll then have a fully-configured yet blank SAN, so people won't be able to work for another 1-2 days until data migration is done.
I have a server (Xserve 2009) and 2 RAID arrays that are currently dark and will run the XSAN. I therefore plan to:
- Mount my SAN backup to the server[/*]
- Direct-attach the XSAN arrays - 1 in each fiber port[/*]
- Install the new server, bind it to my OD, set it as a primary XSAN MDC with no other members at this time[/*]
- Migrate all data from the backup[/*]
- Then I'll shut down everyone and turn over the environment (uninstall MetaSAN, put the XSAN on the fabric, update switch zoning, setup XSAN on clients, setup standby MDC, AFP resharing, etc.) This should be doable in one evening to the degree that users can come the next morning and work.[/*]
- Mount my MetaSAN array locally on ONE machine and synchronize changes since the backup.[/*]
1. What says you? Any caveats I'm missing?
2. The most precarious step in my above plan is the last, since data hasn't been backed up between steps 1 and 6. A MetaSAN volume without MetaSAN is just plain HFS+, but can anything happen if I mount it alongside XSAN? For safety I wouldn't even trust zoning, I'd direct-attach it to a dual-port HBA (no quad-port cards in house, unfortunately).
3. XSAN installs always have pairs of data LUNs. Is this an actual requirement of XSAN, or a function of active/active RAID architecture? Why do more LUNs give better performance? My main data pool (16x2TB) will have to live on a single-controller chassis, so I was planning one big RAID6 LUN. I'd rather have one RAID6 array where any 2 drives can fail instead of 1 drive per RAID.
4. My metadata will live on an older but dual-controler RAID. If I have 4 x 500GB drives available for metadata, what RAID configuration would you recommend for them? RAID10 with metadata & journal on the same volume (and controller affinity=1)? Two separate RAID1's for metadata and journaling?
5. If the secondary MDC is the OD master, what is the power-up sequence?
6. In case I need to roll back, does binding XSAN to an OD server ("Use an existing Open Directory server") "do something" to the directory or modify records in any way?
7. When configuring the MDC, would you point to the OD master by its IP or DNS address?