Growing aggregates vs when to create new ones? (storage philosophy)

[kodiak_f]

Hello,

We're adding additional DS212C shelves to our FAS8200 cluster.  The cluster has 2 non-root aggregates, that are identical 18 drive hybrids w/ a raid group for SSDs and a raid group for FSAS drives.

The cluster has until now been solely for VMWare NFS datastore storage, but we'll be provisioning some general purpose CIFS and NFS SVMs soon.

The performance capacity is really underutilized at this point, our VMWare doesn't churn a whole lot, and our general purpose CIFS/NFS shares have historically been relatively idle as well.

I could see an argument for maxing out the two current aggregates based on increasing the spindle count and getting better performance.

I could see an argument for creating a new aggregate instead and separating the major roles from eachother, the current two for VMWare use, and the new 3rd aggregate for general purpose NAS.

Any opinions?  Couldn't find much on this in the ONTAP docs, probably because there might be conflicting philosophies and no right answer.  Or maybe there is a single right answer and I just don't have the storage experience to know that yet.  Either way, I'd love to hear opinions.

Thanks!

[Mjizzini · Accepted solution]

Hello,

Both options can be done. Below are some ideas to be thinking about before you decide.

• when you add new disks to the old aggregates, the new disks will be used for the majority of the new write calls.
• When the same 2 aggregates is using all shelves on the system, loosing one shelve will offline all the data on them.
• When you create new aggregates, you may need to reorganize the data on both aggr.

That been said, i think creating new aggregates on the DS212C shelves may be a better option in this case.  

[paul_stejskal]

I'd also discuss with your account team. There isn't anything published because each situation is different and each workload is different. Some customers may have an application that is more CPU intensive than disk intensive (heavy deletes, lots of metadata calls to the same files, etc.), so splitting up the workload across two controllers makes more sense. Anything disk heavy would be better with a larger aggregate to parallelize the disk i/os more.