(network) raid question

[oikjn]

forget NR0.  Its a stripe, your data evenly split on ALL nodes.  It appears you are thinking the LUN is assigned to a single node, but that just isn't the case.  Its very hard to find a situation where NR0 is a good idea at all.  Yes, they exist, but 99.9999% of the time its NR10.  Its such a rule that if you make a NR0 LUN just to test you will see that CMC will constantly pester you to change it.

With NR10, your can can do firmware upgrades and almost anything you want and maintain 100% LUN availability with ZERO downtime.  Each node will actually reboot during these maintenance periods, but because of the NR10 structure the servers don't care one bit.  HOWEVER, if you use NR0, any time ANY node is rebooted or loses connection or has any availability problem, your LUN is instantly taken offline.  

I get your question about the inefficencies of the storage with exchange and it being unneeded and you are probably RIGHT...  that said, ask yourself why is the data on the SAN at all?  The nice advantage with the new exchange server is that they say you can now use local disks instead of SAN.  IMO, unless you are running a huge exchange instance with many servers, I would just as soon keep the data on the SAN in NR10 simply because I know the data is going to be available all the time... actually I would probably see about making sure one of the DAG member storage is NOT on the SAN so if for some reason hell freezes over and the san goes down, exchange doesn't stop.

[GilPhilbert]

Johndoew,

What you've implemented means that you have six copies of the data (three mailbox servers x two copies of the data) which are then mirrored at the storage level, totalling 12 copies of the data and excellent resilience, since you can lose a P4000 node and multiple Exchange Mailbox servers and still maintain availability.

However, if you change your volumes to Network RAID (NWR) 0, you'll actually remove all of your storage resilience meaning that the loss of a single P4000 storage node would result in the loss of every Exchange volume and, therefore, all of your Mailbox servers. This is because in NWR0 your volume isn't located on a single storage system, it's striped (spread) across every node in the system.

Think of NWR10 this way: imagine you're out for a camping trip. There's four of you going and you all need a tent to sleep in. With NWR10, you've got two identical tents and each person holds a part of each tent. If one person gets eaten by a passing Griffin, you've still got all the parts to make up a single tent:

• Person 1: Rods / Floor
• Person 2: Floor / Inner
• Person 3: Inner / Outer
• Person 4: Outer / Rods

If Person 2 gets eaten, for example, you've still got a floor and inner to make up a complete tent.

Extending my admittedly odd metaphor, in NWR0 you'd all share the same tent, each carrying a single part of the tent.

• Person 1: Rods
• Person 2: Floor
• Person 3: Inner
• Person 4: Outer

Now of Person 2 gets eaten, you can't put your tent together and you're stuffed.

In my example, the tent is the volume. With NWR0 we have a single copy while with NWR10 we have two copies - that's basically the difference. Most people get mixed up assuming each volume sits on one node (NWR0) or two nodes (NWR10) but that's NOT the case.

As the others have said, NWR0 isn't an option and neither is NWR5 (mainly due to performance). That pretty much leaves you with NWR10.

I hope that helps rather than muddies the waters...