Re: [linux-usb-devel] USB bridge cables in non-network mode

2003-06-13 Thread Bennett Todd
2003-06-12T17:21:01 Greg Brigley:
 Thanks,  I wasn't aware that winpcap could send as well as receive.

Very, very good point, I'd completely forgotten about that, libpcap
is just the sniffer half.

Its complement is libnet, and it seems to have Windows support as
well.

Somehow I forget that this sort of stuff requires two separate
packages.

In the general case, I don't think a great case could be made for
using libpcap+libnet to implement networking tools for normal
production use, rather than using the OS stack.

But your case isn't the general one; you've got this simple cable
with two endpoints and you want a single application to talk over
it. In some abstract sense, using the OS's whole networking stack is
arguably cleaner, but it's understandable that you don't want to
force your users to have to configure networking.

I think that using libpcap+libnet together with the standard
network-interface low-level drivers is a cleaner solution than
attempting to write and maintain a new custom USB protocol
implementation.

-Bennett


pgp0.pgp
Description: PGP signature


Re: [linux-usb-devel] USB bridge cables in non-network mode

2003-06-12 Thread Bennett Todd
2003-06-12T13:20:38 Greg Brigley:
 :) Yes.  From an implementation perspective, using the network
 sounds simpler and more flexible.  But going without it reduces
 the amount of configuration our users have to do, and the amount
 of support that we have to do.

I just occurred to me, any possibility it might make sense to use a
network protocol over the network bridge cable, and save the user
from having to configure Windows networking by building your
Windows-side app on the Window pcap library? Given you've got a pure
point-to-point connection, with only two participants, one of which
can always be the initiator, the full complexity of TCP is probably
not necessary. Writing a simple UDP-based protocol with
acknowlegements, checksums, sequence numbers, and retries should be
fairly painless I'd think.

Don't even worry about address negotiation, just hardwire in one
address for the host side and another for the client; since you
aren't going to be pointing the OS's network stack at this interface
no need to worry about conflicts, no?

-Bennett


pgp0.pgp
Description: PGP signature