Compression of large files

Compression of large files

[Atila Romero]

During a file write, if the compression doesn't proves to be economical
to a portion of the file, the whole file is marked as nocompress, and
subsequent writes don't use compression at all. This can be seen as a 
feature, as it saves CPU in files like videos. But if the file is a dd
image of a disk, or other types of large files, this behavior is
undesirable, since different portions of the file will have different
compression ratios.
I only became aware of this after searching the source code, after
spending some hours trying to find out why the compression wasn't
working. This can be a really unexpected behavior for a end user who is
looking for compression.
The specific part of the code is at inode.c:
477 <#l477>
                /* flag the file so we don't compress in the future */
478 <#l478>                 btrfs_set_flag(inode, NOCOMPRESS);

How about commenting line 478?

Atila

Re: Compression of large files

[Chris Mason]

On Wed, May 27, 2009 at 11:10:09AM -0300, Atila Romero wrote:
> During a file write, if the compression doesn't proves to be economical
> to a portion of the file, the whole file is marked as nocompress, and
> subsequent writes don't use compression at all. This can be seen as a 
> feature, as it saves CPU in files like videos. But if the file is a dd
> image of a disk, or other types of large files, this behavior is
> undesirable, since different portions of the file will have different
> compression ratios.
> I only became aware of this after searching the source code, after
> spending some hours trying to find out why the compression wasn't
> working. This can be a really unexpected behavior for a end user who is
> looking for compression.
> The specific part of the code is at inode.c:
> 477 <#l477>
>                 /* flag the file so we don't compress in the future */
> 478 <#l478>                 btrfs_set_flag(inode, NOCOMPRESS);
> 
> How about commenting line 478?

The plan is to add a flag for the file to make it always compress even
when parts of the file do not compress well.

-chris