by Wolfgang Fischer
New broadband deployments are frequently justified primarily by today’s applications rather than anticipated demands. For example, streaming video content is considered by many as the ultimate bandwidth-hungry application. When one adds the bandwidth requirements of one high-definition TV stream, a few standard-definition streams, and Internet browsing, it may seem that 20 to 25 Mbit/s of bandwidth is sufficient in the long term. But historical data and projections indicate exponential long-term growth in bandwidth demand. Indeed, some service providers are already offering 1 Gbit/s access to residential customers today, and there are substantial deployments of 100 Mbit/s networks in some European countries. These bit rates can only be provided via FTTH.
Certain applications are especially demanding of high peak bandwidth:
- Download of large video files for browsing, editing, or post-processing.
- Joint video editing or other forms of remote collaboration on huge files.
- Telepresence, which includes concurrent video, voice, and application traffic.
Most of these applications have highly bursty traffic patterns and require high bit rates only for a relatively small fraction of time. They can share common aggregation and backbone networks, which can be highly oversubscribed. In contrast, streaming applications such as broadcast video, video-on-demand, or voice over IP (VoIP) require bandwidth to be reserved for the entire duration of the application.
Another factor to consider is the increasingly symmetric nature of the traffic patterns. Peer-to-peer file sharing, e-mail, remote collaboration, VoIP, and others create inherently symmetric traffic streams in contrast to the highly asymmetric client-server type applications like video streaming or web browsing.
Aggregation and backbone networks can be upgraded comparatively easily, and bandwidth increases can be accommodated without much additional investment. Investments into an access infrastructure, however, have to be considered very static. Therefore, network planners must take care to determine whether a seemingly cost-effective access technology will constitute a bottleneck for future bit-rate requirements.
Constructing an FTTH network is very labour-intensive and therefore costly. Experience tells us that the dominant part of FTTH deployment expense goes into civil works, with a relatively small part related to the actual optical cables. This implies that when civil works have to be carried out, it does not really matter how much fibre is being deployed.
Furthermore, while the electronic elements in the FTTH network have a lifecycle of few years, the fibre plant has a longer lifecycle of at least 30 years. This longevity and the high cost of labour required in physical construction places strong demands on proper design of the fibre plant. Once the cable is in place, it is costly to change.
The choice of architecture is therefore important. Star architectures provide dedicated fibres (typically singlemode, single-fibre with 100Base-BX or 1000Base-BX Ethernet transmission) from every endpoint to the point of presence (PoP), where they are terminated on a switch. Endpoints can be single-family residences, apartments, or multidwelling units where a switch in the basement fans out to the apartments using any appropriate transmission technology.
PON architectures for FTTH deployments are characterized by passive optical splitters to distribute the fibre to each customer or basement unit using splitting ratios ranging up to 1:64 or even 1:128.
There are benefits for service providers that deploy PON architectures instead of deploying point-to-point fibres, although some of these turn out to be less compelling than often assumed.
Fibre savings between the optical splitter and the central office or PoP locations is the most relevant aspect for PON FTTH deployments. If a service provider has existing spare fibres or duct space available between the PoP and some street cabinet, then this may prevent the need for digging up the streets.
In a scenario with no existing infrastructure or deployment to new neighbourhoods, fibre savings is usually irrelevant because the marginal cost of additional fibres is negligible when trenches need to be dug, or when right-of-way solutions like sewer pipes can be used.
The fact that there is a dedicated optical interface per customer in a point-to-point scenario may imply that this architecture is inherently more expensive than an architecture sharing ports among a larger number of customers. Experience with a large number of projects, however, has shown that dedicated Ethernet ports are price-competitive given the higher cost of PON ports. Ethernet ports are very inexpensive due to the huge numbers shipped in enterprise and service provider networks, whereas GPON ports are technology-specific and are not yet shipped in significant quantities.
Assuming a 100% take rate of an FTTH offering, the PoP for a PON architecture would have less equipment compared to Ethernet FTTH. But assuming realistic take rates, as discussed later, the difference disappears. This is due to the fact that the first customer on a PON requires an optical line termination (OLT) port, and thus the number of OLT ports cannot be reduced based on a lower take rate.
The management of a large number of fibres appears to be very difficult without the availability of novel optical distribution frames that allow PoPs to be built with several thousands of fibres entering from the outside plant. State-of-the-art equipment can accommodate almost 2,000 fibre terminations in a rack.
Taking these parameters into account, a PoP for 16,000 customers that has been used as an example in the previous issue of Lightwave Europe (see “Going, Going, GPON!” in the Q1 2007 issue, page 11) would require five racks for active equipment using state-of-the-art Ethernet switches (in the realistic case of a maximum 32 customers on a PON tree, a state-of-the-art GPON would require six racks), given a take rate of 30%, which is a very common number for FTTH deployments around the world. The architecture would require five racks for the fibre management. Most service providers would consider this a small premium for the great flexibility they obtain with this approach.
Meanwhile, there are a number of issues facing service providers that deploy PON architectures.
Shared bandwidth. Bandwidth on the PON fibre tree is shared among as many customers as possible in order to benefit from potential cost savings on a per-subscriber basis. As GPON technology provides 2.5 Gbit/s of aggregate downstream capacity, it does not appear to provide for longer-term service growth and future subscriber demands given the exponential growth in bandwidth demand. Furthermore, some proportion of the bandwidth has to be reserved for streaming services, reducing the bandwidth that can be shared statistically.
Encryption. As every PON effectively constitutes a shared medium, encryption is needed on all data streams. However, encryption requires some substantial overhead with each packet that can, depending on the traffic mix, considerably reduce the usable bit rate on a PON.
High operating bit rate. Due to the shared nature of PONs, every endpoint (optical network terminal [ONT]/OLT) has to operate at the aggregate bit rate. Even if a customer has only paid for 100 Mbit/s, each subscriber ONT on that PON tree has to operate at 2.5 Gbit/s (GPON). Operating electronics and optics at 25× the required bit rate has cost implications, particularly where volumes are not very high.
High optical power requirements. Every 1:2 power split causes a degradation of the power budget by 3.4 dB. Consequently, a 1:64 split degrades the power budget by 20.4 dB (equivalent to a power ratio of 110). Therefore, in this model, all the optical transmitters in a PON architecture need to provide 110× more optical power compared to an Ethernet FTTH point-to-point approach given the same distances to overcome.
Local loop unbundling. PON networks do not immediately support local loop unbundling (LLU) requirements because there is only a single fibre connecting a number of customers, which, consequently, cannot be distinguished on a physical level but only on a logical level. This aspect of PON would only support a kind of bit stream offering by the access provider rather than direct subscriber access through LLU. Most of the newer FTTH deployments in Europe explicitly target some form of LLU, which opens up new business opportunities even if not currently mandated by regulators.
Flexibility of the allocation of customers to PON optical splitters can theoretically be achieved by combining the splitter with an optical distribution frame in a field cabinet. This option seems appealing when the take rate of customers is not easily predictable, such as in overbuild scenarios and when regulators enforce LLU requirements. In this latter case, the field cabinet accommodates a splitter per service provider to be served and its associated feeder fibre toward the PoP. However, such flexibility comes at the cost of building and maintaining the optical distribution frame in the field. Every customer change would require a field technician to patch fibres in a street cabinet.
Subscriber reach. Service providers rarely reach a take rate that approaches 100% in residential service deployments. Usually, take rates are closer to 30%, which means that the PON structure is not well utilised and the per-subscriber cost of OLT equipment is increased considerably. One solution is the use of remote optical distribution frames, as discussed in the context of LLU. However, this equipment adds considerable cost that is usually not compensated by the improvement in PON utilisation.
Troubleshooting and maintenance. Any problem in the fibre plant between the splitter and the subscriber ONT is very difficult to locate using standard optical time-domain reflectometry (OTDR) because discontinuities behind the splitter cannot be determined uniquely. This makes troubleshooting very complex in a PON architecture and leads to increased operations cost.
Resilience. An attacker transmitting continuous light into the fibre tree will take down the entire communication for all subscribers on that PON, and it is very difficult to track down the source of this attack.
Technology migration. At some point in the future, the deployed PON equipment will have to be upgraded to new technology with higher bandwidth capabilities. IEEE and ITU-T are working to standardise next-generation, 10 Gbit/s PON requirements. The resulting equipment will not likely be backward-compatible to existing PON deployments (GPON or EPON). For the migration from one PON technology to another one, therefore, all the endpoints have to be replaced.
Ethernet FTTH also provides several advantages when compared with PONs. For example, a direct fibre can provide virtually unlimited bandwidth, which offers the ultimate flexibility for future service deployments as bandwidth needs increase. Typical deployments of Ethernet FTTH access networks use inexpensive single-fibre 100Base-BX or 1000Base-BX technology with a specified maximum reach of 10 km. To support longer distances, there are optical modules available on the market that allow for a higher optical budget.
In addition, only customers that have a subscription with their service provider occupy ports on the Ethernet FTTH access switch. In this case, as the customer base grows, additional Ethernet line cards and customers can be added with a very fine granularity. In the PON architecture, the first customer connected to a tree requires a more expensive OLT port and the associated cost per subscriber is only improved by adding customers to the same PON tree.
Because the singlemode fibre used with Ethernet FTTH is technology and bit-rate neutral, it is easy to upgrade any customer to some higher bit rate without affecting other customers. This means, for example, that a customer who has Fast Ethernet today can be upgraded to Gigabit Ethernet next year just by moving the customer’s fibre to another switch port and by replacing only the Ethernet device on the customer’s premises. All the other customers in the Ethernet FTTH access networks will be unaffected by this change.
Local loop unbundling is an intrinsic property of Ethernet FTTH architectures.
On a physical level, the dedicated fibre is the most secure medium today, particularly compared with any shared medium. In addition, Ethernet switches in service provider environments are purpose built to provide physical port level and logical customer separation capabilities with many robust security features that prevent virtually all infringements.
Finally, though there is little data available that would allow a direct comparison between the architectures, there is evidence that the operational cost for an Ethernet FTTH deployment is less than that for PON FTTH architectures. The table on page 9 provides a summary of the most important aspects.
One aspect not previously discussed is the impact of cable breaks due to construction work. From an Ethernet perspective, the worst case is represented by a break of a large cable with several hundred fibres in the feeder part of the access network. Compared to a cable carrying PON traffic, which typically requires fewer fibres, it will take longer to have this cable repaired. However, using ribbon cables, an entire fibre ribbon can be spliced at a time, reducing the repair time by at least an order of magnitude. Furthermore, this is not different from the copper plant today, where service providers are successfully protecting the main cables leading into their central offices.
Fibre-based access networks are being deployed today based on different architectures and technologies. Mature standards for these technologies and availability of the necessary equipment enable low-risk deployment in service provider networks.
In Europe, deployments are mainly based on Ethernet FTTH in point-to-point topologies with some Ethernet ring deployments from earlier Ethernet service deployments. At this point, PON architectures have made little headway in Europe because most European FTTH projects have been driven by municipalities, utilities, and housing companies. Key factors in many Ethernet FTTH deployments are the flexibility of the business model and the ability to support future services.
Fibre deployment to residences is a huge investment that should last for the next 30 to 40 years. Every deployment scheme for FTTH has its own merits. PONs will have their place in those situations where a sparse fibre environment can save on civil works. However, there is a risk that short-term savings in the fibre infrastructure from PON FTTH deployments will significantly affect the future use of the expensive fibre infrastructure without major follow-on investments.
Wolfgang Fischer is responsible for business development of service provider infrastructures for Cisco in Europe (www.cisco.com).