This isn’t even top 20 in this thread, but here’s mine: maybe 17 years ago, we upgraded our department server from a big old Sun 4/690 running sunos to a shiny

This isn’t even top 20 in this thread, but here’s mine: maybe 17 years ago, we upgraded our department server from a big old Sun 4/690 running sunos to a shiny new Ultra80 running Solaris. Among the many functions this server has was to host a bunch of black and white x terminals. Probably only a few people here ever used those (although more than most other online forums!), but basically the idea is that they plug into the network, at power on they tftp down the image for the x server, they boot and allow you to run x client apps on the server, an 80s thin client implementation. So we upgrade the server and things are working pretty well, especially for such a major upgrade. My boss/mentor at the time is truly brilliant, so we really had most everything thought of. The only thing that was off was that all of a sudden the xterms all stopped booting. We couldn't figure it out. Network sniffers didn't show anything useful--we were just baffled. On a whim, we decided to take the tftp server out of inetd control and truss it (Solaris equivalent of strace). The first time? Worked perfectly--our test xterm booted just fine. Eventually we figured out that the new server was so fast that the speed of the tftp transfer was triggering a problem on the Ethernet card firmware of the xterms and by using truss, it slowed down the transfer and bypassed the bug. Solution: In inetd.conf, we just spawned in.tftpd with "truss -o /dev/null". Never saw the issue again.