2009-02-09 10:46 PM
What are the steps, in general terms, for adding disks to an existing aggregate? I wish to add to performance, so my understanding is some sort of "re-striping" would be necessary to properly re-allocate the extra space?
2009-02-14 08:42 PM
Adding disks to an aggregate is remarkably easy though there are a few things to consider.
First, make sure you've had a quick look at the friendly documentation on this e.g.
This all uses command line for this, though personally most of the time I just use filerview.
In OnTAP 7.0, 7.1 and 7.2 the maximum usable size of an aggregate is about 14TB, and in 7.3 its about 16TB. This means for each disk type, there is a maximum number of spindles that can be allocated to the aggregate. Before you start adding disks check the table below to make sure you arent going to exceed the maximum. For 7.2 and below, the "Drives Per Aggregate" figure is the maximum number of disks used for data and parity, for 7.3 its the maximum number of data disks only. I dont have the figures for the 450GB spindles handy, but if you really need it let me know and I'll dig it up.
You should only add one kind of disk to an aggregate. While its possible to add 300GB disks to an aggregate of 144GB disks, this has some potential performance and capacity issues and is therefore generally discouraged. There is covered in the friendly manuals (see link above)
You should try to add a mimimum of 4 disks to the aggregate at the same time, especially if you are adding disks to an aggregate that is almost full (full that is from the perspective of the write allocation engine). If you dont do this, then all new writes will go to one or two disks which might cause performance problems. Adding four disks at a time gives the write allocator lots of good choices and keep the write efficiency at a good level. Another reason you should always add 4 disks is because all your RAID groups in the aggregate might already be full, so adding new disks will require creating a new RAID group. If you're using RAID-DP (as you should), then adding 4 disks ensures that you'll get at least 2 data disks and 2 parity disks in the RAID group.
If you cant add a minimum of four disks at a time, or you want to maximise the overall write performance immediately, you might want to conside re-organising the phyical layout of the blocks in the aggregate via a re-allocate -A command. For more information see
Most of the time this really isnt neccesary, as with the way OnTAP works, the data will automatically restripe over all in spindles in the aggregate over time as the data ages.
If on the other hand you're adding spindles to increase your read performance for existing data in the aggregate, then running reallocate start -f <volname/filename> may help as the data will be more evenly spread across the spindles in the aggregate, including the disks you've just added. On the other hand, you might also find that adding a PAM card might be a better way to improve read performance. The best way of finding this out is by turning on predictive cache statistics (PCS) (which are available in OnTAP 18.104.22.168 and 7.3 and above), and analysing the results.
2009-04-08 02:39 PM
Thanks for your response John. I have a question. On the link that you referred to it says "After you add disks to an aggregate, run a full reallocation job on each FlexVol volume contained in that aggregate. For information about reallocation, see your Block Access Management Guide."
Does it make more sense to do a reallocate on the entire aggregate at once or is there a reason they specified running the reallocate at the volume level? Will the end result be the same?
Thanks for your help.
2009-04-08 03:39 PM
In most cases it doesn't help at the aggregate level as you'd like your volumes to be laid out better as that's how you'll be accessing it. There are a few cases where an aggregate level may make sense, but Tech Support can advise when that is a good idea.
Keep in mind that you can avoid having to run this at all if you add disks in large enough quantities, IMHO. My personal best practice is to always add disks to aggregates in either full or half of your RAID group (assuming a decent size RG of at least 14 disks, which in RAID-DP is perfectly reasonable). If you do that, you should have enough spindles that new data will have plenty of free space to be written. I'm not saying it's never appropriate if you do this, but it makes it less likly that it'll be necessary. Also avoiding filling your aggregates and the space up-front (say when you hit 90%) is a good thing that will help. But if you need the reallocate, it's there. But most customers I've seen don't need it.
Just some ideas.
2010-03-29 01:11 PM
Great info. You covered all the questions I had except for two very basic ones. The first question is, I'm assuming that adding disks can be done while the aggregate is online and being accessed, yes? The second question is how long can I expect the adding of disks to take? I have a FAS2050. I have an aggregate that has 10 FC 300GB drives and I want to add six. Is this a hour long process or can I expect it to take numerous hours. I have no idea what kind of timeframe this process will take.