Do I need to reserve disk space for UFS snapshots?

[Julian Elischer]

Kris Kennaway wrote:

>On Wed, Apr 26, 2006 at 03:23:39PM -0500, Eric Anderson wrote:
>  
>
>>Kent Ho wrote:
>>    
>>
>>>Hi,
>>>
>>>I'm planing to use FreeBSD snapshots.  It seems to be the snapshop image 
>>>is placed on the same file system.  The question is how do I manage or 
>>>reserve space for the snapshots?  Can someone give me some guidelines or 
>>>pointers.
>>>
>>>Here's my netapp it has a separate area allocated for the snapshots, I'm 
>>>keeping 2 weekly snapshots in there(weekly.0 & weekly.1) using around 30G.
>>>
>>>Filesystem          kbytes       used      avail capacity  Mounted on
>>>/vol/vol0/          305052552  272495320   32557232    89% /vol/vol0/
>>>/vol/vol0/.snapshot  76263136   30658028   45605108    40% 
>>>/vol/vol0/.snapshot
>>>
>>>Filesystem               iused      ifree  %iused  Mounted on
>>>/vol/vol0/            10802387   89822586    11%   /vol/vol0/
>>>
>>>If copy all content in vol0 to a FreeBSD and make 2 snapshot will it be 
>>>around the same? 272G + 30G = 302G?  So if I df on the FreeBSD box 
>>>around 302G will be used?
>>>
>>>Any help and information is highly appreciated.
>>>      
>>>
>>FreeBSD snapshots do not work the same way as NetApp snapshots.  First, 
>>they must reside on the same filesystem as the snapshotted filesystem. 
>>They will initially eat up some space upon creation equal to the space 
>>used by the cylinder group summaries and superblocks, but it isn't a 
>>whole lot.  You'll need to have however much space is changed weekly for 
>>each snapshot, so if 10Gb per week is changed, then you'll need that 
>>much space per snapshot file done.
>>    
>>

Kent,

The best way to think of UFS snapshots is to consider that there are two 
filesystems on the same media.
They share a lot of data when they are in sync (just after the snapshot 
is created) so there is very little
overhead.. (think of it as a bit like having all the files linked).
As the files on the live filesystem are changed, the blocks that hold 
the original data are unlinked from
the live filesystem, and replaced by new blocks that hold the new data. 
They continue to be linked
into the snapshot filesystem however. So the overhead grows to be the 
size of the difference
between the live filesystem, and the snapshot.