Debunking the Myths of Shared Networks: The Point-to-Multipoint Effect

Curtis Knittle
Vice President, Wired Technologies Research and Development

Oct 11, 2016

“I don’t want to have to share a pipe. The problem with ‘cable’ is shared pipes. If my neighbor is doing a bunch of stuff over the network, I get impacted too. With fiber I get speed and no shared pipes.”

--- Entrepreneur in a focus group

The notion that subscribers connected to residential fiber networks do not “share pipes” is often misunderstood. For residential fiber networks, sharing pipes is one of the main reasons fiber to the home (FTTH) is even remotely cost-effective for service providers to deploy. But what is most surprising is the following: deploying shared network solutions has led to a more rapid increase in residential broadband speeds than otherwise would have been the case with non-shared access network solutions. I like to call this the Point-to-Multipoint Effect. In the process, sharing pipes has allowed broadband speed growth to surpass the predicted 50% compounded annual growth rate commonly known as Nielsen’s Law of Internet Bandwidth. Read on to learn more…

First, a couple of definitions:

  • A (non-shared) point-to-point (P2P) network topology is one in which there is a single dedicated connection between two endpoints. In the case of access networks, one endpoint is typically located at the hub or central office, or could be located at a remote distribution point. The other endpoint is a digital subscriber line (DSL) modem, for example, or a simple Ethernet switch, located on the customer premise. In P2P networks, the peak capacity of a link is used exclusively by only the two endpoints.
  • A (shared) point-to-multipoint (P2MP) network topology is one in which there is a single downstream transmitter and multiple access termination devices that all selectively listen to the same downstream data stream. A key characteristic with P2MP networks is the peak capacity of the network is shared between all connected endpoints. Two examples of P2MP networks are HFC and passive optical networking (PON), shown in the figure below (showing downstream transmission).


Two examples of (shared) point-to-multipoint networks: HFC and PON

The PON solution represents the most prevalent residential fiber solution in the world, primarily due to lower costs compared to P2P fiber solutions. To illustrate the sharing, referring to the diagram above, if 10G-EPON is the technology choice, each optical network unit (ONU) connected to the network transmits upstream at ~10 Gbps, but they don’t transmit simultaneously. Instead, an ONU must be scheduled by the OLT for upstream transmission to avoid collisions with other ONUs. In essence, the scheduling of ONUs results in the sharing of the 10 Gbps peak capacity.  Consequently, there is a whole lotta pipe sharing going on in PON solutions.

Do shared networks necessarily perform better or worse than non-shared networks? It depends on how performance is measured, but in one key area, residential broadband speeds, shared networks have significantly outperformed non-shared networks by a substantial amount.

A recent blog discussed Nielsen’s Law of Internet Bandwidth and how the cable industry was preparing to meet future broadband speeds with 100G-EPON. When Mr. Nielsen made his initial prediction in 1998, residential broadband access was dominated by dialup and ISDN connections, which are both P2P solutions. Indeed, for approximately the first 14 years since that initial 300 bits per second dialup connection in 1982/1983, the progression of available peak service tier bit rates followed the 50% annual growth rate prediction.

**various sources compiled by CableLabs

The release of the first DOCSIS® specifications by CableLabs in 1996 essentially represented the dawn of P2MP solutions, i.e. shared, for residential Internet connectivity. According to the data in the chart above, the tremendous rate of technology advancements resulting from the shared DOCSIS/HFC network solution, and later with the development of shared PON technologies, coupled with the relative cost-effectiveness of these solutions, has far exceeded other P2P technologies for residential broadband. While the initial growth prediction in 1998 was a 50% annual growth rate, the Point-to-Multipoint Effect increased the growth rate closer to 70% for residential Internet connectivity. The Point-to-Multipoint Effect indicates that sharing pipes for residential connectivity has provided a solution that has actually allowed residential high speed data rates to increase at a faster pace! This “sharing” trend is expected to continue with the development of Full Duplex DOCSIS and 100G-EPON, making the introduction of new services possible.  Thus, just like our parents always told us, it is good to share.

In his role as Vice President Wired Technologies at CableLabs, Curtis Knittle leads the activities which focus on cable operator integration of optical technologies in access networks. Curtis is also Chair of the 100G-EPON (IEEE 802.3ca) Task Force.


Keeping Pace with Nielsen’s Law

Curtis Knittle
Vice President, Wired Technologies Research and Development

Sep 15, 2016

The telecommunications industry typically uses Nielsen’s Law of Internet Bandwidth to represent historical broadband Internet speeds and to forecast future broadband Internet speeds. Mr. Nielsen predicted many years ago the high-end user’s downstream connection speed grows by approximately 50% compound annual growth rate (CAGR). In reality, actual peak service tiers offered by service providers over the years may be following something closer to 60% compound annual growth rate, as shown in the figure below.


The point of this blog is not to debate whether the growth rate is 50% or 60%, but rather if the growth rate continues, how do we evolve our networks to keep pace?

For point-to-multipoint networks there is a general rule of thumb for determining the peak service tier given a particular peak network capacity. This capacity-to-peak-tier ratio of 2:1 isn’t necessarily based in scientific fact, but comes from years of experience that a 2:1 ratio allows service providers to have a reasonable level of confidence that speed test measurements will accurately reflect a user’s subscription level. For example, for a particular access network technology, if the network supports 2 Gbps transmission rates to/from the access termination device (i.e., a cable modem) then the peak service tier typically won’t exceed 1 Gbps.

The present state of the art access network technology peaks at 10 Gbps[1]. The IEEE 802.3 10 Gbps Ethernet Passive Optical Network (10G-EPON) has been deployed in China and the United States. ITU-T has recently consented XGS-PON, another 10 Gbps symmetric PON standard that uses the physical layer of XG-PON (ITU-T G.987.2) and 10G-EPON. Even the ITU-T’s NG-PON2 standard, which uses multiple wavelengths to increase network capacity, only defines a single wavelength per optical network unit (ONU), which puts NG-PON2 on par with 10G-EPON and XGS-PON in terms of meeting peak service tier rates. Finally, CableLabs is now certifying DOCSIS 3.1 devices which are capable of 10 Gbps downstream, and soon will certify 10 Gbps symmetric devices based on Full Duplex DOCSIS technology. What does this mean for accommodating Nielsen’s Law? Assuming the peak service tier trends continue, and given the 10 Gbps peak network capacity of current solutions, the maximum peak service tier will level off at approximately 5 Gbps (see red dashed line in chart above) until technology advances to allow higher rates. The telecommunications industry needs a technology roadmap beyond the current state of the art which allows for peak service tiers to exceed 5 Gbps.

CableLabs and its members, along with other service providers and the IEEE, are determined to stay ahead of the trends displayed in the graph above by contributing to the world’s first 100 Gbps EPON solution as part of the IEEE 802.3ca Task Force. The prevailing sentiment of the 802.3ca Task Force is to create a generational standard that allows for growth of peak network capacity (and corresponding peak service tiers) if and when such growth becomes necessary, without creating a new standard. This growth is expected to be achieved through defining four wavelengths, with each wavelength supporting 25 Gbps. Initial product developments will revolve around a single wavelength to provide a 25 Gbps EPON solution. When market conditions demand it, using two wavelengths along with a channel bonding solution will allow an ONU to transmit and receive at up to 50 Gbps. Similarly, with four wavelengths and channel bonding the ONU will transmit and receive at up to 100 Gbps. Examining the chart above, and assuming historical trends continue, the reader can see the 100G-EPON standard will support peak service tiers out to approximately 2030, give or take a couple years, assuming the 50% CAGR predicted by Nielsen continues.

One of the interesting facets of the 802.3ca Task Force activities relates to the improvement in efficiencies in the media access control (MAC). Previously, the IEEE 802.3 standard did not allow frame fragmentation, but recently with the completion of the IEEE 802.3br Interspersing Express Traffic Task Force, frame fragmentation is now allowed in networks based on the 802.3 standard. The 802.3ca Task Force plans to leverage fragmentation to make transmission more efficient in a multi-wavelength, channel-bonded environment. Additionally, contributions to the 802.3ca Task Force will improve the efficiency of the upstream bandwidth allocation process by allowing multiple service flow queue depth reporting and upstream granting in a single message pair. Considering the ITU-T SG15/Q2 is also investigating 25 Gbps per wavelength, the more promising and exciting aspect of these 802.3ca Task Force decisions is that the next generation of IEEE EPON and ITU-T GPON standards could be more closely aligned than ever before in the very near future! This makes a converged optical access solution closer to reality. (see a previous blog regarding a converged optical access initiative)

In his role as Vice President Wired Technologies at CableLabs, Curtis Knittle leads the activities which focus on cable operator integration of optical technologies in access networks. Curtis is also Chair of the IEEE 802.3ca Task Force.

I’m ignoring point-to-point fiber solutions such as Active Ethernet because these are not typically residential solutions.