Re: [TUHS] ARPAnrt, and the Unix epoch

Yes, it was a string of changes the network did on January 1. A few years earlier (1980?) they switched to long leaders on the IMPs on Jan 1. This Jan 1 cutover meant that Mike Muuss and his staff (including me) at BRL always had work to do over the holidays. Shutting off NCP support was easy. All NCP control messages traveled on the IMP link 0. IP traveled on link 155. The IMPs were a gentle test for IP. The virtual circuits were reliable and flowed controlled even if the early TCPs were ornery. A good IP implementation did try to deal with the RFNM throttling. The only real issue is the MTU mismatch between Ethernet and the IMP (1008 vs. 1500). A while later, I noticed that the BRL Gateway (an early IP router) was printing out messages that it was receiving link 0 messages. Someone had unblocked link 0, and there was some site out there was still some NCP host trying to contact us (we had moved all our actual hosts OFF the outside imps in favor of a pair of BRL Gateways). Our interior network at that point also comprised a handful of our own class B addressed IMPs. An amusing story there was as they were delivered and uncrated by the BBN guys I was running around connecting up the data trunks and plugging hosts into them. I sent a message to our BBN contact and told them that I had done this and I got a long message back explaining it was impossible. I'm sure glad I didn't know it was impossible when I did it. Later on, Joe Pistritto managed to get the NOC protocol spec out of someone, and we wrote our own NOC for the private IMP network. Still, when they finally split the Arpanet from the Milnet, we got them to do it on the first of the Fiscal Year (October) which fit our time schedule at the labs a lot better. Amusingly, there was one small hitch with the "MAILBRIDGE" gateways that connected the two networks. Apparently, BBN forgot that we had the only IP router that was connected to the MILNET (other than the MAILBRIDGES) and they screwed up that routing.