2013-11-12 03:46 AM
We have a new FAS2240-2 and by default it seems to have taken 50/50 ownership of the disks and I wanted to know if this was best practice as I assume if we only had a single controller then all would be assigned to that one.
My main issue is that even after configuring the HA cluster it seems that they are seperate, i.e. each has an aggr0 and vol0 and only 8 disks can be assigned to a new aggregate per controller.
I would have thought that all the disks could be assigned to a single aggregate or have I got the wrong end of the stick here?
Sorry but I know with DELL Equallogics with 2 controllers all the disks in the array are available together.
Any help clearing this up would be really appreciated.
Solved! SEE THE SOLUTION
2013-11-12 03:49 AM
Each controller in HA pair is independent system and manages own disks. HA partners do not have access to each other disks unless failover happens and one controller goes down.
So what you see is correct.
2013-11-12 07:19 AM
Aborzenkov is of course 100% right, but may be I can elaborate a bit.
The way I explain to colleagues with backgrounds using lesser storage, is with an HA NetApp you're getting two autonomous storage devices which are capable of taking over each others disks in a failure situation.
You don't have to split it 50/50 but each head will normally require a minimum of 4 disks (1 for data, 2 parity and a spare). If your need is for a single large area you could assign 20 disks to one head. But usually with a bit of thought you can split your data more evenly.
2013-11-13 02:25 AM
Thank you both for your replies but nigelg1965's elaboration is really the answer I was looking for, so thanks again for that.
I have used NetApp in the past quite some years ago but with many more shelves so the split of disks wasn't as obvious, its just this unit is a effectively a single shelf with 24 disks and having 2 small 4TB aggregates for what is a fairly powerful storage system seems a little bit wasteful of disks due to each controller requiring the 4 disks to operate.
I will leave the split as it is so my next question would be do I make more effective use of space by extending aggr0 where I can get around 6TB or create a new aggr1 with around 4TB. Are there any reasons that could cause any headaches or limitations by have one big aggr0 in the future which could be avoided by going with a new aggr1 now ?
Thanks in advance.
2013-11-13 03:18 AM
There are two schools of thought here, one says how a separate aggregate for vol0 and the other says create one large aggregate per filer (within the limits of aggregate size / disk type / size). I'm definitely of the latter school, especially with so far disks relatively. Creating another aggregate per head is going to "waste" 4 disks and unless you going to be adding more shelves very soon could actually make disk access slower.
2013-11-13 03:52 AM
In this case then I very much tend to agree with you, one aggregate it is then, makes sense to me as I think it may be quite some time before we go for another shelf.
Really appreciate your help.
2013-11-13 03:58 AM
To offer my opinion/experience..
1 - Splitting vol0 onto an independent aggregate doesn't seem to give much benefit on smaller filers (or larger ones for that matter). I see it written in multiple places, and from multiple sources, that it is best practice but I have only ever been able to find 2 reasons to implement this:
2 - If you do have an independent aggr for vol0, you can always use RAID4, so you only require 2 disks and your spare can be shared with the data aggregate. This is a great strategy for smaller filers with a single disk shelf.
3 - If you are not needing the filer performance you gain by splitting your data disks across 2 heads, then I would look at assigning all the data disks to a single head and use the other one solely for fail-over. We do this where we can only justify a smaller filer (2000 series) with a single disk shelf, but obviously want to maximise the amount of storage we get from our investment. In this case, you would have a 2-disk aggregate for vol0 on the second head (no spare, as it is only for fail-over) and the rest of the disks assigned to the first head and put into one large aggregate (hosting vol0) with 1 spare disk. This gives the benefit of more spindles in your data aggregate (which will give more performance than splitting across 2 heads) and you don't have to waste 2-3 disks (2 parity+1 spare) by creating an aggregate on the second filer. The downside is that you have to take the risk of the RAID4 vol0 with no spare on the 2nd head but this is only there for fail-over and if you are really risk averse then you can always add a spare there. Either way you will have saved some parity disks and gained performance by having more spindles in your raid-group..
On larger filers, we tend to run separate vol0 aggregates on production and obviously share disk shelves between heads, but for smaller filers with 1-2 shelves, 1 head is more than capable of serving everything..