cck@CUNIXC.COLUMBIA.EDU (Charlie C. Kim) (01/01/88)
As I have been expermenting, I have found two things I want to comment on. First, a nit-pick. When running in EtherTalk mode, DDP checksumming seems to be turned off. This, to me, sounds like a really bad idea. Depending on the Ethernet CRC only tells you that it reached from hardware to hardware correctly as far as the hardware can determine. There are conditions where hardware errors may result in incorrect packets being sent and received. The problem is compounded when one is attempting to send over "long-haul" networks. c.f. recent commentary on mod.protocols.tcp-ip (aka tcpip mailing list) re the use/non-use of checksumming for NFS. Second, I've seen a condition where an AARP request packet is sent before most if not all packets sent to a particular unix machine. After some work, it was isolated to one of three items: (a) ethernet address, (b) minimum packet size, and (c) pecularities of a particular ethernet interface. After further testing, it was isolated to particular types of ethernet addresses. The particular machines are DECNET hosts and so have an ethernet address of the form "AA-00-04-00-xx-04 where some combination of the upper level bits tell it is a decnet address and some combination of the lower bits tell what the decnet address is. This certainly sounds like a probable "problem" in EtherTalk 1.0 as distributed with the current Apple EtherTalk hardware. Charlie C. Kim User Services Columbia University