ietf-822
[Top] [All Lists]

Re: Transport Level Compression?

1998-09-13 09:56:16
So, while there may be value in a compressed C-T-E, it is
something I'd expect to be applied when two communicating MTAs
decide it is appropriate for us with any data that met some
observable criteria (probably having more to do with the
patterns of the bits than the semantics of the file).  As soon
as you start talking about transferring precompressed objects of
particular sorts, I think you are back to an application/
subtype, not a C-T-E.

I'm afraid I'd have to disagree. Without getting into whether or not a
compression scheme independent of a data distribution format (and remember that
this is a separate proposal from the FS data distribution format) such as this
is a Good Thing, if you want to do one it pretty much has to be a CTE.

Why? Simply put, because nothing else works. The problem with overloading the
syntax of content type and subtype is that these fields already contain useful
information that agents need -- they tell you what the underlying data is and
how it is to be handled. Define a subtype for "compression" and you lose the
fact that the underlying data is, say, an Excel spreadsheet.

In fact there is language in RFC 2047 that specifically forbids this sort of
content type definition, for precisely this reason.

It has also been suggested that this would be best done at the transport
layer. Unfortunately this isn't the right place either, for two reasons:

(1) Compression is expensive, and you don't want to have to redo it after every
    hop. (There was also a suggestion that intermediaries would simply store
    the compressed form. I'm sorry, but this dog doesn't hunt, as email
    transfer necessary requires modification of the transferred content at
    every hop.

Reading some of your other notes, my guess --perhaps completely
wrong-- is that your more general model/plan wants to see a
hierarchy of the form of
    C-T-E / content-type / content-subtype / parameters
That is probably unfortunate, at least from the standpoint of
those of us who see C-T-E and content-type as orthogonal, not
hierarchical.

I don't think this is the intent at all. I think the intent is to define a
separate compression component. And yes, the problem with this is that
compression is in general tied to the type of data you're dealing with. But
this doesn't mean a general compression layer cannot be made to work and would
not be a Good Thing to have.

                                Ned


<Prev in Thread] Current Thread [Next in Thread>