mike@BRL-TGR.ARPA (07/18/84)
From: Mike Muuss <mike@BRL-TGR.ARPA> Error threshold switches should be UP (set to 15) if your ECU--ECU link is faster than 19.2K; for slower links you might wish to use a threshold of about 4. Owning ECUs is silly if you don't use their error correction feature, and setting the error threshold to zero (all down) disables it. The theory here is that it's better for your ECU's to do a retransmission locally wherever possible, rather than stomping on the packet and having TCP do an end-to-end retransmission across the network. (MILNET and ARPANET trunks are only 56 Kbps; best to conserve their bandwidth when it's easy to do so). As for the override switch on the ECU, the setting of this switch really depends on two factors: whether you are interested in hearing about hard-retransmission errors, and whether your host is doing RFNM counting. If your host is doing RFNM counting, this switch MUST be down, or your network service will get worse, every time a RFNM from the IMP to your host is dropped. (VAX 4.2 UNIX does do RFNM counting, I have changes which make it selectable). With the switch down, you will also get an imp0: interface reset message (4.2 UNIX) every time you get a hard error on your ECU link. In addition to clearing the RFNM counters, this also gives you a record of how bad your ECU--ECU line is. If your host is not doing RFNM counting and does not log IMP interface resets, you might as well leave the switch up. (Some dumb gateways fit this description). Best, -Mike PS: We use a 480,000 bps modem between our main gateway and our MILNET IMP, so that no matter how noisy the line gets, we can always be sure that the ECU's are not the limiting factor in our gateway's performance.