You are correct, my assumption is that there is a socket-like layer for any
type of=20
connection that we connect a printer to and that IPP rides on top of this=20
layer. This assumption is true for TCP/IP and for parallel connection with=
=20
1284.4 support. If we encounter a connection, such as serial where that is=
=20
not the case, then we must define a similar layer to implement sockets,=
which=20
hopefully borrows a lot from 1284.4. [Question to 1284.4 experts: could it=
=20
work with serial or USB with very few changes?]=20
In effect, I want the socket layer to handle the issues of multiplexing and=
=20
read/write blocking and let the IPP layer concentrate on printer operations=
=20
transmitted over a virtual connection.
Bob Herriot
At 04:52 PM 5/6/98 , you wrote:
>One thing it would buy is a simpler (than 1284.4) way to flow IPP over bidi
>parallel - no? Isn't this basically what Lexmark has found? I'm confused=
why
>TIPSI has a packet structure, Lexmark was shipping it on parallel and then=
.4
>was invented - maybe some background could help (I always thought it was to
>flow SNMP over parallel ;-). If it's true that the .1 packet is only still
>there for legacy I might buy Bob's argument. But I suspect .4 is a much=
more
>complex implementation.
>
>Bob, doesn't your proposal say we would have to invent a transport (if not
>already there) to "IPize" every physical layer (ex. serial)?
>
>Harry Lewis - IBM Printing Systems
>
>
>
>owner-ipp@pwg.org on 05/06/98 05:32:34 PM
>Please respond to owner-ipp@pwg.org
>To: sdp@pwg.org
>cc: ipp@pwg.org
>Subject: SDP,=A0 IPP>PRO Proposal for TIPSI-like protocol
>
>
>I just finished scanning IEEE 1284.3 and IEEE1284.4.=A0 The most=
interesting
>part is Chapter 8 "Service Provider Interface (SPI)" in IEEE 1284.4.=A0=
This
>chapter describes a "Berkeley Sockets-compatible interface for clients and
>servers to access the services provided by 1284.4".
>
>So if I understand the intent of 1284.4, it is to provide a layer that
>supports sockets over parallel connections. All we need to do in IPP is
>reference sockets for TCP/IP and 1284.4 and we don't have to worry about=
the
>issues at that layer.
>
>So, I conclude that we don't need to packetize IPP or do much of what is
>proposed in Roger and Harry's paper. Instead, we can send IPP directly on
>sockets layered on top of TCP/IP or 1284.4.=A0 There are a few=
easy-to-solve
>dangling
>issues, such as chunking for document data and intermediate acknowledgement
>when attributes are verified for PrintJob. But otherwise IPP stays as is.
>
>If you disagree with my conclusions, I would like to know what the
>TIPSI-like packetizing layer provides that sockets don't also provide?
>
>Bob Herriot
>=20