>>> 2020-10-02 so net (PDF)

Only my being very busy has spared you from an excessively lengthy discussion of historic long-distance telephone carriers. Let me give you an extremely brief summary:

Long-distance telephone calls were initially carried on "open-wire" carriers which were not radically dissimilar to the analog twisted pair used in the local loop (e.g. to your house). These carriers had limited capacity (one call per open wire pair) and call quality tended to deteriorate noticeable with distance. The next generation in long-distance carrier technology was analog multiplexing, in which frequency division muxing (and occasionally other methods) were used to place multiple analog telephone calls on the same wire. These were primarily transmitted via coaxial cables or microwave radio. Next up, digital carriers took over, in which audio was digitized and time division muxing was used to multiplex calls onto a single carrier, which was once again coaxial cable or microwave. Soon, though, digital carriers moved over to optical fiber, which is the primary media today.

There's a variety of different ways to pack bits across an optical fiber, including Ethernet. Some of them are described with words like "Plesiochronous." Perhaps the most beloved of the telephone industry, though, is synchronous optical networking or SONET. SONET is directly motivated by the need to carry ISDN-type data channels (e.g. "DS" channels which are carried over the "T carriers" we talked about earlier), and so there are many design similarities.

The most complicated part of SONET has to do with synchronization---in order to do time division muxing well, everyone needs to agree on exactly when each time slot starts, and in a distributed system this can become quite complex. Perhaps fortunately for you, though, I honestly don't find the synchronization component to be that interesting, and so I pay a lot more attention to the logical payload of DS0, which is fixed-sized frames which are allocated to different data channels using traffic engineering methods (that is, in advance).

Ironically, considering that SONET is all about being Synchronous, outside of telephone calls SONET is most commonly used to carry ATM, or Asynchronous Transfer Mode. ATM is a rather simple protocol which was developed as part of the ISDN standards. Being a telephone protocol, ATM is logically connection-oriented. This means that a connection is established using a a circuit establishment protocol, and ATM frames are then routed based on knowledge of the connection that they belong to. Somewhat like TCP, this creates the illusion of a fixed end-to-end connection on top of a packet-switched (i.e. non-connection-oriented) network. But, because connections are explicitly established in advance (rather than being an 'emergent property' of the protocol), ATM is able to make significantly more assurances about quality and reliability of connections.

The desire for a high degree of consistency in performance leads to some interesting design decisions. Perhaps most prominently, ATM packets (called cells) are of fixed size. Messages smaller than a cell will be padded to make them the standard length. While this is inefficient in terms of bandwidth, it simplifies switching, queuing, and other operations, and most importantly leads to very consistent and predictable performance since all cells should take identical processing time and resources. The fixed-sized cells are also particularly amenable to transport in other protocols that use fixed-size data units, such as SONET.

ATM is also quite nonspecific as to the contents of cells. Much like MPLS, this gives ATM a degree of "protocol Independence" that makes it a popular "lowest common denominator" used in networks that transport heterogeneous protocols. When I say "heterogeneous protocols," I don't mean HTTP and HTTPS. I mean packet-switched, variable-bandwidth protocols like IP and circuit-switched, fixed-bandwidth protocols like BRI telephone calls---and all with traffic engineering facilities that prevent the opportunistic protocols impairing the fixed-bandwidth ones.

Much like ISDN promised, SONET with ATM ought to be a good choice to carry telephone, internet, and perhaps other digital services (virtual private ethernet, anyone?) over a single connection. It became very popular for this purpose, and was widely used by larger businesses for their internet and telephone connectivity.

What I find most interesting about SONET though is the physical architecture. SONET was typically deployed using "add-drop multiplexers," which sit "midway" down a SONET link and operate by removing the frames from some time slots (and making them "incoming traffic") and replacing them with new frames (the "outgoing traffic"). This switcheroo provides symmetric inbound and outbound bandwidth, but oddly in different directions. Incoming traffic comes from the left, and leaves to the right, for example. For this reason, SONET is deployed in "loops" that typically run from the telco through multiple clients and back to the telco.

Further, although exact arrangements vary, loops generally had a "clockwise" fiber and a "counterclockwise" fiber, and often redundant fibers in each direction as well. Depending on the type of deployment, the two directions might be used for increased bandwidth (with redundant pairs for backup) or simply for backup, with the ability to send traffic "the other way around" if the connection is interrupted at some point in the loop. These loops typically had multiple clients on them, and it seems to have been a common practice to equip the add/drop multiplexers at each client site with a network diagram showing which clients were next in each direction---perhaps to aid troubleshooting when the connection drops. You can always call your neighbor and ask if their network closet is on fire. I once interned at a facility where one side of the SONET cable went to an intelligence agency, and we used to chuckle about the nature of their traffic running through our SONET equipment. The other side went to an astronomical observatory, and that might be enough to figure out who that employer was if you're clever.

I enjoy this particular aspect of SONET because it seems to introduce this odd concept of neighborliness into network architectures, and it's a pleasing variation from the star topology that is the norm in last-mile internet connectivity. Of course, ultimately, SONET was not able to deliver the best-case data rates that strictly packet-switched systems could (e.g. Ethernet), even if it delivered a much better worst-case. SONET is dying out in favor of "native IP" solutions like wide area ethernet (think very long ethernet cables). As we speak, networks are getting more boring.

Next, and hopefully in a shorter interval than has been typical lately, I will talk a bit about the internet delivery technologies available to consumers today, like DSL and DOCSIS. I will also share some opinions about wireless internet service, because you know I have some of those. I gave a talk at a conference of sorts about this topic so I can easily recycle material, and I always love doing that.

I also have a probably more interesting but closely related topic I wanted to talk a bit about: cellular phones, and the evolution of cellular networks. I'm working on a personal project right now that touches this area heavily and jeez is it making me wish for the old days of analog carphones.

Incidentally, while it has been occupying quite a bit of time, my landings are becoming less jarring by the day and I hope to fly an airplane solo soon, perhaps next week. Maybe I'll be able to post more often once I'm spending less time googling "how to hit ground gently."