On Thu, Apr 26, 2001 at 12:06:52AM -0700, Aaron Voisine wrote:
> > All the data on freenet is inserted after going through a round of
> > encryption and thus having the protocol do any compression is a waste of
> > resources becuase compression of the encrypted stream in most cases won't
> > compress it.
> >
> > -Mathew
>
> er... yeah... I knew the data was encrypted. I didn't realize that the
> encryption rendered the data mostly random. I suppose that makes sense,
> recognizable patterns can't be good for security. I guess the only way to
> do it then would be to compress that data prior to encryption. This would
> have the added benefit of reducing the size of file in datastores as well.
> Of course this would not be a backwards compatible change. It would still
> be beneficial to have freenet nodes automagically zlib compress files
> prior to insertion. The other problem would be that the algorithm can't be
> changed in the future without breaking freenet.
>
> What if the first part of the unencrypted file were:
> Compression=zlib:
> Followed by the compressed data. This way "none" could be chosen for
> already compressed files and new algorithms could be added in the future.
> If the file does not start with Compression=<algorithm>:, it could be
> assumed that this file was inserted prior to the compression feature,
> allowing for backwards compatibility.
One thing you could do is create a new command in the metadata standard to
specify that the data section is compressed with algorithm X.
The other way we could go is rip Metadata-length out of storables and instead
make it a mandatory client convention to include a FieldSet at the beginning
of each document giving the metadata length and other optional flags like
Compression-zlib ... of course FCP and the reference clients would hide
that stuff from the user.
--
# tavin cole
#
# "The process of scientific discovery is, in effect,
# a continual flight from wonder."
# - Albert Einstein
_______________________________________________
Devl mailing list
[EMAIL PROTECTED]
http://lists.freenetproject.org/mailman/listinfo/devl
