Request for advice configuring/mounting XRAID

Question

I'm asking for some advice about configuring Xserve RAID.
I am not an IT person, but just a researcher and have found
myself in charge of some computing resources.
We have an Xserve RAID - 14 drive, each set of 7 on RAID5 - connected
through fibre switch to an Xserve head node. (The fibre switch is
because we will probably add another Xserve RAID in future)
The intent for the RAID is to hold our data while we do our computations.
We have a small cluster of Xserve cluster nodes for distributed Xgrid and
Matlab computation. I intended to network mount the RAID on each computer
in cluster. All the users will ssh in to a shell, and none will login to
a desktop.
I have just begun to setup the RAID, but ran into some questions that
I thought I would ask for advice before I proceeded.
I have already striped both controllers together with software RAID.
(diskutil createRAID stripe "Journaled HFS+" "RAIDvol" disk2 disk3)
It created the stripe and created the jHFS+ file system, but I didn't
realize it would also mount it in /Volumes/RAIDvol
1. Do people have advice on whether I should keep this mount point?
I had rather thought I would mount in root (/) on head node and
cluster nodes, e.g. /RAIDvol instead of /Volumes/RAIDvol
(And how should I change the mount point? Evidently not through fstab
- should I use NetInfo(niutil) and is that complicated?)
Before I change it from mounting in /Volumes, are there any reasons not to?
(again, no one is going to see a desktop, everyone will ssh into a shell)
2. Should I have striped the disks together, or concatenated them?
(What's the difference? - any performance difference?)
3. Do people have advice on network mounting, AFP vs NFS? What
performance differences should I consider?
Thanks for everyone's help.
dmaus
Xserve RAID   Mac OS X (10.4)  
  Other OS  

Answer 1

>1. Do people have advice on whether I should keep this mount point?
By default all external volumes will mount under /Volumes.
It is possible to change, but I'd leave it there, it's not worth the hassle trying to bury it somewhere else, and if no one's logging into the desktop then no one's going to notice it anyway. Since it's being shared to the other clients it can be mounted at any point outside of /Volumes, but I'd leave it in /Volumes on the head node.
Trying to change this will just run into problems later on when Apple change their model.
>2. Should I have striped the disks together, or concatenated them?
(What's the difference? - any performance difference?)
Striping is better than concatenation.
The difference is in how the volume is laid out, and specifically how the data is written to the volumes.
If you write a large contiguous file to a striped volume, a chunk of data (32K by default) is written to the first LUN (half of the array) then the second chunk us written to the second LUN). This alternating pattern gives you the best throughput since the RAID is effectively writing to al 14 disks at once, and can be flushing the buffers on the first 32K while the next 32K of data is being written to the other LUN.
By comparison, concatenation simply appends the second LUN. As you write your file it is all written to the first LUN and it's only when this LUN is full that you start writing to the second LUN. Therefore your throughput is much lower since you're only using half the disks at once.
Concatenation is easier to setup since some volume formats allow you to grow a volume by simply adding more space at the end, whereas slicing requires a reformat, but those are your choices.
>3. Do people have advice on network mounting, AFP vs NFS? What
performance differences should I consider?
Performance is one factor, security is another. NFS doesn't offer any significant security other than which IP addresses are allowed to mount the volume. AFP supports authentication to ensure that the specific user can mount the volume.
This may or may not be relevant to you depending on how the clients are mounting the volumes, and whether you want there to be any crossover between the clients. If they're being mounted at startup (so the volume is always available) I'd probably opt for NFS. Home directories, on the other hand, would be better served via AFP.
Performance wise NFS might have a slight edge, but I've heard others say AFP does, so YMMV.