Aggregate space

[kevinmajenta]

Is there a best practice for the amount of free space to be left within an aggregate?

Regards

Kevin

[radek_kubka · Accepted solution]

Yep.

At the end of the day, aggregate is just a file system, so if it gets fuller, the performance drops.

Two rules of thumb: never go beyond 90% full & ideally do not fill over 80%.

Of course your mileage may vary & if, say, you rely heavily on thin provisioning across many volumes then obviously keep us much free space in your aggregate as humanly possible!

Regards,
Radek

[pascalduk]

radek.kubka wrote:
 
Two rules of thumb: never go beyond 90% full & ideally do not fill over 80%.

It depends on how you define usage. Filling an aggregate 100% with almost empty fully guaranteed volumes does not have an impact

Your calculations have to be on the actual data stored in the volumes. The sum of all data in the volumes preferably should not exceed 80-90% of the aggregate space.

[radek_kubka]

 Filling an aggregate 100% with almost empty fully guaranteed volumes does not have an impact

According to my best knowledge that's actually not the case - ONTAP is not allowed to use the area within guaranteed volumes for any swapping, maintenance, etc., so performance will struggle if aggregate gets 'nominally' close to 100% full.

Also WAFL panic may occur when there is no free space left in the aggregate, which is even worse (basically the filer hangs)

Regards,
Radek

[eric_barlier]

In my opinion I d say up to 90% tops. Opinions vary as you can see.

If you are looking for more space you should assess not having aggr. snapshots. Our shared service model does not allow for an entire aggr. to be restored so we re-claimed the aggr. snapshot space.

Eric

[amiller_1]

Just curious....what did you set your aggregate snapshot reserve too?

[eric_barlier]

Hi Andrew,

We use 0%, we backup using snapvault and/or snapmirror.

Eric

[amiller_1]

I've actually gone with a best practice of 3% given the notes around problems with upgrades to 7.3 if less than that is free (using aggr snap reserve to guarantee that) and then just keeping a single daily aggregate snapshot online (even if don't want to use that, rolling back to that and then pulling differences from snapmirror would be faster than a full restore).

If you don't mind, what all did you base setting aggr snap reserve to 0% on?

[pascalduk]

If you don't mind, what all did you base setting aggr snap reserve to 0% on?

The likelihood of restoring the complete aggreagte to a snapshot is extremely unlikely.

[eric_barlier]

Hi guys,

yes for us in our config. it would be way too intrusive to revert using an aggr. snapshot. IMHO its like rebuilding a house in order to fix a broken window.

Cheers,

Eric