KCM@BYUADMIN.BITNET (Kelly McDonald) (04/11/88)
It appears that this is becoming an important issue since there appears
to be at least four different implementations of IP over netbios.
1. My KA9Q driver for netbios.
2. The implementation at Waterloo.
3. the Cornell tn3270 implementation.
4. CMU driver done at Univ. of Wisc.
To my knowledge, 1 and 3 use netbios datagram support to transport the
IP packets to the gateway. I think 2 has both datagram or connection
support,(is that right?) and 4 uses netbios connections.
My personal opinion was that datagrams were adequate, though we had to
live with the 512 byte limit. I believe that we should try and merge
support to some extent so that there would be a compatibility of
products. We chose the KA9Q base because it offered SMTP and FTP service
to the PC lan, and are using it as the gateway between the campus ethernet
and the departmental lans. However, we would certainly like to use the
Cornell tn3270 product for IBM host access. It appears that the only
difference in the implementation has been the choice of the netbios
datagram name.
In any case, I'm interested in a serious discussion to merge our direction
on this subject.ljm@TWG.COM (Leo J McLaughlin) (01/05/89)
Some time ago I posted a specification for IP over NetBIOS. The primary concern of the responses I have recieved is the use of broadcast datagrams instead of NetBIOS group names for IP broadcasts. The problem with using NetBIOS group names for broadcast is that it isn't quite as simple as just saying 'use IP.xx.xx.xx.xx'. For instance, if I have a class 'B' address that uses 24 bits of subnetting I need to post 7 listens: IP.81.54.4.2, IP.FF.FF.FF.FF, IP.81.54.FF.FF, IP.81.54.4.FF, IP.00.00.00.00, IP.81.54.00.00, and IP.81.54.4.00. Worse, this pretty much prohibits NetBIOS hosts from being multihomed (each logical interface using 7 listens), requiring the host software to monitor the link layer interfaces chosen before allowing IP address assignment. (Latest IP broadcast info taken from the December 25th edition of the Requirements for Internet Hosts specification). My belief at the time was that more NetBIOS applications would mind the stealing of all those NCBs than would mind the use of broadcast datagrams. We tested our implementation with a handful of IBM programs (the most important of which was PC LAN) and all of them worked. So I ask for anyone who knows of some common NetBIOS applications which do require the non-use of broadcast datagrams (or for that matter of any NetBIOS implementations and/or applications which would mind if many listens were posted) to please send me mail. leo j mclaughlin iii The Wollongong Group ljm@twg.com