FreBSD 9.1 and ZFS v28 performances

[Ronald Klop]

On Wed, 20 Mar 2013 13:45:01 +0100, <kpneal at pobox.com> wrote:

> On Wed, Mar 20, 2013 at 01:32:19PM +0100, Ronald Klop wrote:
>> On Wed, 20 Mar 2013 11:45:37 +0100, Daniel Kalchev <daniel at digsys.bg>
>> wrote:
>> >> Well, after changing the recordsite property, I copied the file from  
>> an
>> >> UFS partition (using cp -Rp): this should use recordsize=16k, right?
>> >
>> > Perhaps, if you delete the file, or preferably the entire ZFS dataset
>> > first. Copying an file over another existing, does not change anything
>> > with the destination file except it's contents and modification times.
>> > As is always with changing settings, it is safer to just create the
>> > entire data set from scratch, with the new settings.
>> >
>> > Daniel
>>
>> ZFS never overwrites contents of a files. It always allocates new  
>> blocks.
>
> True but not really relevant.
>
> Taking an existing file, truncating it to length zero, and then putting
> data into it results in the same file having different contents. But
> deleting the existing file, creating a new file, and putting data into it
> gives you (like I said) a new/different file. This is true with both UFS
> and ZFS.
>
> Applications don't care that ZFS does COW under the hood. Applications
> care that the observed behavior of ZFS be similar to UFS.
>

It is relevant. After changing the recordsize all new blocks will get the  
new recordsize.
The discussion was not about if a file is the same one or not. It is about  
if the recordsize changes. And recreating the volume/pool is not needed  
for that.

Ronald.