56K really was the maximum for telephone modems. This is because the telephone network, internally, transmitted 8 bit samples at 8K samples/sec - 64Kbps - except the occasional low bit got stolen for other signaling, so the net data rate was a bit less (and the result was inaudible to a telephone customer). Instead of trying to synchronize the modem to the stolen bits, it was simpler to just use the 7 that were reliable - thus 56Kbps. https://en.wikipedia.org/wiki/Robbed-bit_signaling
9600bps as the defacto terminal speed was older than 9600bps modems (those were the cat's pajamas in the mid-late 80s or so). It was simply "fast enough" for a terminal user and "slow enough" for the terminals of the day. I had one (a Volker-Craig 4404 if I remember correctly) that could not keep up at 9600bps and had to be used at something less, maybe 4800bps.
The reason 9600bps was The Speed for terminals is that with 10-bit framing (start/byte/stop) you could send an entire 80×24 screen in two seconds, while not overloading the (fractional-MIPS) host with interrupts or DMA when doing so for many users.
Typically a minicomputer terminal multiplexer batched transmission when possible, so it only had to issue an “all done” interrupt to the CPU when it was done sending a batch of characters it pulled from memory via DMA. Similarly, in character mode it only had to issue a “something’s waiting” interrupt to the CPU when a receive buffer was full or a sequencing character (like carriage return) was received.
This is also why so many mainframes and minicomputers used block mode terminals: Applications use a forms API and all editing/filling happens on the terminal: The computer transmits a full form to the terminal, the user fills in the form including correcting mistakes and possibly even sees some validation (eg numeric versus alphanumeric fields), the field contents are sent to the computer upon an explicit “done” action.
Even for “fully” interactive use many such systems preferred to use a line mode, where the current line contents are only sent upon specific actions, and line editing takes place on the terminal. UNIX style “raw” mode was too high-overhead for the number of users that systems needed to service, even if the terminals cost a bit more.
Even with a 56k modem until broadband became widely available many of us were stuck with a 28.8 connection, rural phone lines could rarely handle 56 reliably.
I heard Shoemaker or Levy once talk about the modem on Voyager...running at a blistering 10 bps. I keep our old 2400 bps modem around as a curio. The memories are interesting and fun for me, but I get that they may not be so much for some of the other engineers I work with.
My first modem was 1200 baud! This was back in 1988, I think.
I remember when I upgraded to 9600 in the early 90's. It was an incredible upgrade. I also ran my first SLIP connection on that machine (an Amiga), probably around 1993 or 94.
The 4800bps was directly attached to a computer, no modem. If both were set to 9600, the terminal would miss characters. But most dumb terminals were fast enough to keep up at 9600.
56K really was the maximum for telephone modems. This is because the telephone network, internally, transmitted 8 bit samples at 8K samples/sec - 64Kbps - except the occasional low bit got stolen for other signaling, so the net data rate was a bit less (and the result was inaudible to a telephone customer). Instead of trying to synchronize the modem to the stolen bits, it was simpler to just use the 7 that were reliable - thus 56Kbps. https://en.wikipedia.org/wiki/Robbed-bit_signaling
9600bps as the defacto terminal speed was older than 9600bps modems (those were the cat's pajamas in the mid-late 80s or so). It was simply "fast enough" for a terminal user and "slow enough" for the terminals of the day. I had one (a Volker-Craig 4404 if I remember correctly) that could not keep up at 9600bps and had to be used at something less, maybe 4800bps.