Inception - 2000
In they year 2000, I landed my first job working at a business that also operated a dial-up ISP. At the time, the state of the art was 56k dialup. At speeds below 56k, a symmetric arrangement of modems and telephone network was possible. The subscriber had a modem and the ISP had a modem, connected through the analogue telephone network in the middle. Of course, the telephone network was only analogue for the last mile, everything in between was digital. But this double conversion, analogue to digital, and back again, capped modem speeds at 33.6 kbps. 56k technology was made possible by requiring the ISP segment of the connection to be digital as well.
This meant that to run an ISP with a 56k connection, you had to have a digital connection to the telephone network, a Primary Rate Interface, otherwise known as a DS1 or T1 connection. A traditional T1 runs over two pairs of wires; one for transmit and one for received, and has a maximum range of around 6600 feet. In order to maximize cable pair usage, and minimize the number of repeaters, a number of technologies were developed to attempt to reduce the outside plant burden of provisioning a T1 circuit.
In the 2000’s, Digital Subscriber Line rose to popularity. Most folks are familiar with DSL as a way to get a high speed internet connection. The same technology can be used in the telephone network to deliver T1 circuits to subscribers. Typically, high speed internet connections use Asymmetric DSL, or ADSL, which has more bandwidth in one direction. A T1 connection requires a symmetric bandwidth profile, which is achieved by allocating an equal number of channels in both directions for communications.
The ISP I worked at had a number of HDSL2 circuits in place: they were used for both the PRI circuits and for the connection to the internet. Having only dial-up at home, the technology fascinated me.
Phase 1 - 2011
11 years having passed since first discovering HDSL, equipment could now be had cheap enough on eBay that I could make my dream come true. I managed to scrounge up a couple Cisco routers, WIC modules, and most importantly, a CO/remote pair of ADC HDSL2 modems. The HDSL2 hardware arrangement isn’t perfectly symmetrical, the central office unit (with C in its model number) provides high voltage DC to power the remote unit (with R in its model number), and the CO unit is also in a form factor more convenient for installation in a CO.
There are a few different CO form factors available:
- H2TU-C-231 - 220 mechanics
- H2TU-C-388 - DDM+ mechanics
- H2TU-C-319 - 3192 mecahnics
- H2TU-C-202 - 200 mechanics
For phase 1, I used H2TU-C-231 modules, because they were the cheapest ones I could find on eBay at the time. What I didn’t realize at the time is that the card edge connector used for those cards was a non-standard size, and impossible to find. It took a bit of sweet talking to convince a sales rep at Sullins to send me a single sample.
The remote units all have the same form factor, 200 mechanics. I used a H2TU-R-402 for my tests. I ultimately ended up with this arrangement:
With all of the pieces assembled, I was able to set up a working T1 span. 1.544 Mbps bidirectional communication:
This project was an interesting reminder about serialization delay. We forget these days about pesky things like serialization delay. At 10 Gbps, a full MTU ethernet frame (1500 bytes) can fit inside a 250 metre segment of fibre optic cable. The transmitter will have finished sending the frame before the receiver is even aware it exists. It takes 1.2 µS to send a full ethernet frame at 10 Gbps.
At 1.544 Mbps, the serialization delay is noticeable. A full MTU ethernet frame takes 7.8 mS to transmit. As I started to mess around with fractional T1’s, I was surprised how much more latency I would see on a ping test as I cranked down the bit rate. Of course this is all obvious, it takes 188 mS to transmit a packet at 64 kbps, so the minimum round trip time for a ping is going to be 376 mS, even though the devices are right next to each other. It makes sense, but in an age where the usual source of latency is distance, it took me a moment to wrap my head around it. With modern networks, 376mS is a very reasonable round trip time to the other side of the planet.
Phase 2 - 2016