Will Packet Rule Forever?
Musings on the Evolution of Telecom Networks
Try this experiment. Call your friend on WhatsApp Voice when they are sitting across the room from you. Done, what did you find? Yes, wasn’t that a rather large delay from hearing his or her voice across the room to what you hear on your phone. The reason is that since the call is packetized, it uses long buffers to smooth out jitter that may dramatically reduce the voice quality as it traverses multiple router hops on the public Internet.
Kind of weird. But it got me thinking about a Big Question. How did telecommunication networks get to the way they are today, and how might they change in the future?
|50 Years Ago||Today||Future|
|Predominant Application||Voice||Best effort data||AR/VR real-time experience|
|Access||Fixed||Mobile and Fixed||Mobile dominant|
|Transport||Electrical - expensive||Optical – approaching zero cost per bit||Optical – effectively zero cost per bit|
|Information packaging||TDM||Packet||Packet with on-demand TDM overaly|
A scant fifty years ago, telecom networks were entirely different. They were engineered to transport analog voice, using fixed access only, and electrical single channel transmission. In a huge contrast, today’s networks are engineered to transport digital data (carrying infinitely more information, with voice simply as an application), using both mobile and fixed access, and multi-channel optical transmission.
While many technologies contributed to enabling this revolution, one clearly stands head and shoulders above all others, VLSI. When Very Large Scale Integrated Circuits hit the 10,000 transistors per chip mark in the early 1970s, everything started to change.
This first impact was digitalizing voice and central office switches, which brought large energy savings. However, that was just a baby step. The sustainable impacts of VLSI on telecom were:
- Giving rise to ubiquitous computing and the need to network computers with each other, and later on with cloud-based applications.
- Giving rise to a technology curve of ASICs and DSPs that enabled exploiting radio and optical electromagnetic spectrum with frightening year over year gains in efficiency.
Let’s look at the second point first. VLSI density has increased over six orders of magnitude in the last 50 years, from the 10K transistors per chip mentioned above, to over 30 billion today.
That correlates almost exactly with the evolution from 2G digital radio (we can ignore 1G analog radio) from max rates of 40Kbps, to 5G radio rates of up to 30Gbps.
Optical has benefited even more. Its rates have increased over seven orders of magnitude (perhaps with a boost from laser technologies) from 45M over a 10Km span, to over 600G for 100Km. Optical also gained an additional 100X improvement from broadband amplifier technologies that made dense wavelength division multiplexing economically feasible.
Except for some legacy last mile copper-electrical connections, VLSI has brought us to a telecom network dominated by radio (access) and optical (access and backbone) transmission.
So much for transmission. The other transformation occurred in how we “package” information for transmission. Human evolution dictates that voice communications likes low latency jitter free continuity. This is why networks were originally built using TDM circuit-switching technology, where latency and jitter can be controlled precisely.
However and now we come back to point #1 above, computer communications considered TDM a waste of bandwidth, because it devoted time slots to a communications session whether they were filled or not. Packet networking using statistical multiplexing was the way to go. Whether a packet arrived a few 10s or even a few 100s of milliseconds later had little impact on data communications. Moreover, packet routers were able to use frames, which simultaneously handled small to long strings of packets, with very little impact of efficiency.
For a while, TDM and packet technology lived side by side. However, as data communications began to dominate (global mobile data surpassed mobile voice traffic in 2009) packet switching inexorably started pushing TDM into the legacy category.
Which brings us to today and our experiment with WhatsApp Voice. The question now is whether packet-networking ubiquity has rotated too far. AR/VR is beginning to give rise to user-experience applications where not only bandwidth, but latency also matters. Indeed, 5G defines a specific URLLC (ultra reliable low latency communications) mode. The current packet network simply does not have the tools to handle this, particularly with any degree of assurance.
As a result, we are beginning to see TDM approaches such as FlexE and TSN (time sensitive networking) creeping back into the networking mix. This is early days in understanding how best to use and exploit the benefits of TDM in a packet world, but the needle is swinging back a bit, and it will be interesting to see how far it goes.