Path: utzoo!utgpu!water!watmath!clyde!cbosgd!ucbvax!MCGILL1.BITNET!ROBERT From: ROBERT@MCGILL1.BITNET (Robert Craig) Newsgroups: comp.protocols.tcp-ip.ibmpc Subject: Re: TCP/Novell gateway Message-ID: <8801232106.aa04561@Louie.UDEL.EDU> Date: 24 Jan 88 01:42:46 GMT References: Sender: daemon@ucbvax.BERKELEY.EDU Organization: The Internet Lines: 50 Sorry about the earlier "empty" reply to this message, my fingers got away from me. >We are very interested in gatewaying our Novell PC networks to a >TCP/IP Ethernet. > >One unique approach to this problem would be using a PC as a >level-3 gateway between the Novell IPX net and the IP Ethernet. >Client programs (such as NCSA telnet, Phil Karn's NET package, X >windows programs, etc.) would be modified to encapsulate the IP >packets in IPX packets. The gateway would catch the IPX packets >from the clients, strip the IPX headers, and send them out over >the Ethernet. If you're going to use a PC anyway to connect the two networks and act as a gateway, you don't need to do the encapsulation. The two protocols will coexist quite nicely on the same physical network. The only ugly part is that getting NetWare and whatever TCP/IP product you use to share the same network interface is unlikely at best, so you either boot up with the NetWare shell OR with TCP/IP... I suppose one could arbitrate usage of the single interface, but it would be tricky. If you're dead set on the idea of encapsulation, then why bother with the gateway PC? Just encapsulate the IP packets in IPX and pass them through the NetWare bridge to a machine on the backbone ethernet which strips the encapsulation and forwards as appropriate. Don't forget that this machine will have to understand BOTH routing protocols! Or write a NetWare Value Added Process which carries out this function in the file server (bridge) itself. I considered this approach briefly in a moment of insanity. It would mean convincing the bridge to pass packets it didn't understand on to your process so that you could fix them up. It would also mean some rather tricksy coding in a box that one would like to be REAL reliable. If you're going to put the function in the file server, you're probably further ahead to make the VAP talk to another network interface (connected to the same netware network) over which it has exclusive control. This would also allow the use of TCP/IP, rather than this clumsy encapsulation of IP in IPX. This is, I believe, the approach taken by Micom/Interlan and also by what's-their-names. Robert Craig McGill University Computing Centre