>56kbps connections were roughly symmetrical

I don't think so, asymmetry was the innovation that made 56K possible on POTS (plain old telephone sevice, with only enough bandwidth for squawky voice)

From a spec standpoint, acoustic modems were building over wires spec'd for bidirectional voice transmissions, i.e. symmetrical.

I'm curious about the nuances, but it seems like the last mile download/upload imbalance was created by the originating signal mode?

Download = First mile internet to ISP could be upgraded to digital, and thus grab some extra throughout by avoiding analog noise handling

Upload = First mile user to ISP was inherently analog over phone lines, and so sacrificed throughput for line noise tolerance

https://www.edn.com/an-introduction-to-the-v-90-56k-modem/

It was 56k down, 33.6k up.

Sorta-kinda.

v.90 didn't work at all between two regular POTS-connected analog modems. In order for a v.90 connection to happen, the ISP-end of the connection needed to be a digital circuit (typically using ISDN PRI).

By being digital, the gear at the ISP-end was able to precisely and distinctly control each individual bits that would ultimately be converted to analog at a point that was physically near to the user (their local CO switch). This was what gave us asymmetric nature of "56k" v.90.

Eventually, we got good enough at learning how to handle changing line conditions and thereby twiddle the bits with a modicum of precision in the upstream direction. This allowed us to produce a standard with a bit more symmetry: v.92.

v.92 offered up to "56k" (~53k due to FCC limits) down, and 48k up.

A lot of users -- at least in the US -- never experienced v.92. It wasn't formalized until right around the turn of the century, which corresponded well with the time when xDSL, DOCSIS, and/or BRI started showing up even in fairly small not-completely-rural communities at fairly reasonable prices. The local dialup ISP market was beginning to die by then and many never bothered with upgrading their gear to support v.92 before they closed their doors for good.

(All of this wacky dial-up modem tech was both enabled and limited by digital switching in the PSTN. Speeds over dry-pair phone lines could be far higher if there wasn't a digital conversion in the middle, and avoiding that digital conversion is how DSL became possible.

Which is interesting: A DSL circuit was meant to go only across town (ish), and was always betwixt two fixed points. But a point-to-point v.90 or v.92 connection could be established to any properly-equipped machine by just dialing its phone number, and that machine could be across town or on the other side of a continent; it didn't care.)

I would think symmetric bandwidth is baked into switched circuit networks by definition.

A bit earlier than 2000 /g https://lineofsightgroup.com/wp-content/uploads/2017/01/phon...

One of the coolest things about the modem age is that you could easily try a different ISP if you weren't happy with your current ISP's performance (upstream and down).

You can still do that in some places that have open fiber networks! I actually have two ISPs right now running over the same fiber connection (via separate PPPoE connections) - one ISP gives you fixed IP addresses with custom reverse-DNS for cheap so I use it for my apex domain, and the other ISP is Dynamic DNS but has much better performance.

The 56k specs were asymmetrical, the downstream connection could be 56k (actually 53k) while the upstream connection was only 33.6k. The key with v.90 was the ISP could send a fully digital signal down to the customer but they could only send an analog signal up which was capped by physics to 33.6k.

The later v.92 spec supported a digital upstream and could hit 48k upload.

Regardless, dialup users did not have a lot of upstream bandwidth available. They also suffered through high packet loss and latency making their throughput even lower than their line speed would suggest.