The Case for EPON
Ethernet passive optical network (EPON) technology is the ideal technology for 3G and especially for 4G backhaul transport. There are now, and will be in the future, a variety of backhaul transport schemes to handle the increasing variety of cellular backhaul transport. EPON is well suited for femtocell, microcell, and cellular backhaul.
The tree architecture of a PON is the best network architecture for the high density of cells required for 4G technologies. Some 4G technologies will have a higher density of cells using smaller high frequency cells. All 4G technologies will have a higher density of cells to support higher numbers of subscribers and greater data rates. All of the 4G technologies also feature support for Ethernet as the primary base station network interface.
EPON is the only PON technology that supports transparent Ethernet transport compliant with the Metro Ethernet Forum standards. It can transport the access from one or a few users on a femtocell within a residential fiber to the home (FTTH) installation or hundreds of megabits per second to more than a 1 Gbps for multiple-tenant cell sites. As much as Ethernet and EPON are replacing T-1s in the enterprise today, so too will they replace T-1s in cell backhaul.
EPON access systems use deterministic schedulers with guaranteed (not probabilistic) data performance down to 1 kbps granularity. Deterministic schedulers offer worst-case delays that provide low latency and low jitter.
Historically, cellular backhaul has been built on what was the only available transport, time division multiplexing (TDM). In the United States, the cell backhaul market is dominated by T-1 service. The technology, throughput, performance and economics of T-1 service were historically a reasonable match to the technical and business demands for cellular backhaul.
Of course, it is no accident that T-1 is a good match. Cellular systems were in part designed around T-1 since it was the only universally available transport technology nationwide. T-1s are the gasoline and oil of cell backhaul: necessary and addictive. It will take creativity to engineer a replacement.
Changes in both technical and business demands have brought cellular operators out in the open looking for new backhaul transport schemes. The current (and some of the future) needs of cellular operators are well documented.
One of the most fundamental changes in the business has been the transition from usage-based models to tiered and now unlimited use models for voice and more recently for data. During the same period of time, cellular networks have become more geographically pervasive, roaming has become transparent, and capacities have continued to increase in a mature and predictable manner. Overall wireless networks have become a reliable substitute for primary line telephony service.
Many residential subscribers have abandoned their primary (wire) line in favor of wireless primary line service. The same is beginning to happen for wireless data. Wireless data capabilities have grown and become more pervasive and cost effective. Now some subscribers are replacing their broadband access and Wi-Fi services with flat-rate higher performance mobile data services.
The use cases for cellular have been affected by these changes. Primary line residential use, for example, differs significantly from the demands of peak-time urgent business use. A three-hour conversation between teenagers is a very different use than a quick call to say the meeting was cancelled. Even more demanding are the usage patterns for data. Both on-phone mobile data and mobile data access for computers are driving wireless cellular demands.
For a long time, T-1 (and fractional T-1) was the only widely available telecom transport service in the United States. Although integrated services digital network (ISDN) and Switched 56 offered alternatives on the lowest end of the capacity scale, T-1 (DS-1) products have been the mainstay of the telecommunications business. With the growing needs of voice and data capacity, the T-3 (DS-3) products moved from a carrier transport to a customer provided access service as well.
Both products are viewed with awe and admiration as the most reliable service to emulate. In reality, those services are no more reliable than any other tariffed public switched telephone network (PSTN) services. Another way of saying this is that they generally have been low reliability products. Statistics available at the FCC indicate that for years the PSTN has had approximately "two-and-a-half nines" of uptime, or about 99.5 percent. This is consistent with many years of experience with these services.
While the providers of these services claim these figures, there were literally no guarantees or service level agreements (SLAs) behind them. Neither the tariffs nor commercial contracts provided anything even close to the old Bellcore spec of "four nines" availability. Cellular operators know this. In fact, they count on and plan around backhaul failures. Every solid cellular network design has an added layer of network redundancy in overlapping tower coverage. Additional layers include over-engineering cellular power budgets and – even further – roaming agreements in their own native territory as a last effort redundancy.
As a TDM service, it is true that T-1s have both a low latency and extremely low jitter. 2.5G cellular systems designed to operate in that environment have little or no buffering and are therefore intolerant of delay and variable delay (jitter) beyond the ANSI T-1X1 specifications.
It is possible to engineer PON access systems with equally low jitter, but they must also be deployed and configured correctly to do so. In contrast, there is no configuration option for T-1s that would accidentally provide intolerable jitter. Although the capabilities of T-1 exceed the needs of cellular backhaul, we find that the overall one-way delay budget for tower to WSC service is approximately 8 to 10ms. This could be divided up any arbitrary way. In practice, we have seen requirements such as the three examples in Table 1.
Of course, the most significant consumption of the budget is the data rate-product delay. Of this, the transmission distance is by far the most significant single factor for the vast majority of towers. It is easy to understand that T-1 is appealing because these requirements can easily be met by T-1 service offerings with no engineering effort and no risk of misconfiguration or other variations. But it is not true that only T-1 can provide for these requirements. It is also true that as cellular operators’ demand has grown, they can no longer afford to continue to purchase numerous T-1s.
PON offers a lower cost and higher capacity substitute. For 2.5G systems, there is a tradeoff for that reduced cost. That tradeoff is the introduction of pseudo-wire transport equipment that emulates the T-1 transport service to provide physical T-1 interfaces for the 2.5G equipment. The capex and opex of that new equipment, along with the mild risks, are the tradeoffs for lower opex (backhaul) costs and greater backhaul capacity. It is certainly a decision each cellular operator must make carefully.
On the upside, pseudo-wire systems offer significant buffering capabilities that relax the transport jitter requirements. As they transition to 3G and certainly at 4G, most and all base stations, respectively, will find native IP/Ethernet interfaces and will no longer require T-1 interfaces and the micromanagement of access transport latency.
With literally millions of customers running voice and video over the same converged transport and access networks, it’s not a stretch to believe that the quality requirements for cable operator voice over Internet protocol (VoIP) and video are just as stringent as that for cellular, and that EPON access on cable networks offers adequately low latency for cellular voice and data traffic.
Finally, some theorists argue that T-1 is more reliable because of the synchronous optical network (SONET) transport capability typically used for multiple T-1 services. Although SONET can certainly be configured with (UPSR, BLSR, or linear 1:1 or 1+1) protection, the fact is that the vast majority of cell sites served by SONET transport are single-fiber-fed and have no redundancy. Those few systems with collapsed loops offer the appearance of redundancy, but are subject to single fiber cut failures that we have personally seen on many occasions with SONET services.
There is no doubt that providing a redundant ring or linear transport with diverse fiber feeds would increase redundancy and contribute to increased reliability for both fiber cuts and some types of system failures.
The good news is that diverse fiber feeds are possible with PON just as they are with SONET. We’ll discuss the options for diverse fiber PON solutions in a bit.
Myths about PON
There are as many myths about PON as there are about T-1. Although PONs have been around since the early 1990s, many people believe that PON technologies are new. In truth, they are about as young as cellular itself. PONs are widely deployed in general. EPON specifically is the lead PON technology, with more than 12 million ports deployed globally.
Like all PON flavors, EPON is a passive network in the outside plant. This makes it a highly reliable network technology because it is not subject to component or sub-system failures in the outside plant (where they are most vulnerable). EPON access devices are also very cost effective.
PONs are mythically considered expensive. The fact is that a carrier class EPON optical network unit (ONU) device costs less than any T-1 access equipment. PONs are also rumored to be unreliable, either because they are "new technology" or because they are supposed to be built of less than carrier grade technology. EPON technologies are made by many of the same manufacturers as T-1 and other technologies.
Perhaps the best answer to critics is to demonstrate that PON does work, is reliable, and is being deployed for cell backhaul. Verizon Communications is using its FiOS access network to operate cell backhaul for its own (Verizon Wireless) backhaul transport. Verizon was not the first and certainly not the only operator to do so. Later, we will examine the architecture and performance of Bright House Networks’ deployment of EPON for cellular backhaul.
EPON is a mature, reliable and scalable technology that is very cost effective for cellular backhaul. It isn’t just a new access technology; the point-to-multipoint and passive outsi de plant architecture of PON represent dramatic architectural changes and improvements for wireless backhaul.
SONET was initially designed as a core network technology for telcos. It was the ideal technology for inter-office facilities (IOF) where the number of core sites is relatively small, but reliability and multiplexing needs are high. Like SONET, dense wavelength division multiplexing (DWDM), in particular removable add/drop multiplexer (ROADM), systems are designed for a similar set of requirements.
Beyond T-1, SONET became a necessity as one of the few options available for telcos with the desire to provide higher capacity access services. But SONET rings were not designed to support hundreds or even dozens of sites. (See Figure 1.) Although new low cost SONET access equipment helps to reduce costs from the complex multiple-box solutions of the past, the ring architecture is a poor one for distributing and collecting traffic on cell towers that are essentially uniformly distributed.
SONET was also optimized for add-drop multiplexing. That is the carrying of traffic as a virtual mesh – from any one site to any other site on the network. Architects of SONET solved the minimal connector and found rings to be optimal for meshing small numbers of sites with redundancy (two paths from each vertex). But cellular backhaul does not require mesh connectivity.
The theoretical problem of cell backhaul is a connected graph with no cycles. This problem can best be represented and solved as a tree. (See Figure 3.) Most cellular networks actually have more than one WSC for redundancy. In this case, the ideal model is actually a bi-partite graph, which is a special case of a tree with two sets of vertices. The first set is typically composed of at least two WSCs. The second set is the set of all tower sites. When we observe a bi-partite graph (tree), we can readily see the visual similarity to the problem of cell backhaul design. In practice PONs can be designed using passive splitters (that are vertices in the graph) to create a bi-partite logical graph and a more complex physical tree for drop and continue architectures that are best suited to the physical geographic distribution of cell towers.
In short, the architectural challenge is to design a bi-partite graph to overlay on what we call a stick graph in the outside plant world. Both theory and practice show that the cost of fiber consumption is far lower with trees (PON) than with cycles (SONET rings). It is possible to build a bi-partite graph using only linear SONET extensions. Of course, SONET requires fiber for each linear extension. The advantage of PON is that it requires only a single fiber (diplex operator) strand leaving the transport service provider’s facility.
As mentioned previously, most cell towers do not operate with diverse fiber paths. The reason is that the fiber construction cost is prohibitive. If such construction is desired, it is possible to build a redundant solution using EPON access. Such a solution can be as reliable and protection switch as fast (in some cases faster) as SONET protection switching.
The solution is simple for an IP network operator. Two fiber routes are constructed to the tower, one from each of two separate transport operator facilities, exactly as would be done with SONET. From each facility, a PON access network is constructed to the tower. At the cell tower site, the transport operator deploys an IP/multi-protocol label switching (MPLS) U-PE device. This device is an IP router specifically designed for customer access to a redundant IP/MPLS network. Using off-the-shelf EPON small form factor pluggable (SFP) ONUs, the EPON access network is terminated inside an SFP. To the IP/MPLS U-PE router, the EPON ONU SFP appears to be a Gigabit Ethernet (GigE) connection through the gigabit media independent interface (GMII).
The U-PE router can then use off-the-shelf IP/MPLS protection switching techniques (and there are many available) to create primary and second or dual paths for point-to-point Ethernet over MPLS (EoMPLS) or point-to-multiple with virtual private LAN (local area network) service (VPLS) or hierarchical VPLS (H-VPLS) and EoMPLS or Ethernet paths.
The resulting architecture can protection switch with fiber cuts, EPON ONU SFP, port, router, switch, or any transport failure in the order of tens of microseconds at the fastest and tens of milliseconds at the slowest. The actual speed depends on which IP/MPLS protection method is used and the speed of detecting the failure. Using ECI Telecom’s SR series IP routers, with fast bi-directional forwarding detection, and IP/MPLS, we would be able to obtain single digit millisecond protection switching in most cell tower cases. To date, no cellular operators have committed to purchasing a dual-fiber route option because of the additional construction cost. The technology we have offers this capability with no new products required to achieve it.
Bright House Networks operates an all-IP transport network that uses both native IP and IP/MPLS transport for in-house and customer circuit transport. Before deploying EPON for cellular backhaul, the network already used large scale Ethernet transport for business services. The technical requirements for throughput, latency and jitter in the converged transport network are as stringent for voice, video, and data – as they are for any cellular voice or data applications.
Bright House Networks proposed and deployed an EPON access solution in combination with its IP/MPLS network for cellular backhaul for a customer. The design and deployment model for cellular backhaul was targeted at leveraging the existing infrastructure. Both the capital and operational cost savings of using an existing network are superior compared to the buildout of a SONET/T-1 access network.
Performance delay budgets
For 3G, the delay budget was such that EPON provided no specific challenge. As discussed previously, 3G and 4G networks have buffer capabilities specifically designed to support IP/Ethernet transport schemes. The greatest challenge was 2.5G service delivery. The 2.5G systems lack buffers and therefore require tightening the delay requirements beyond the 3G specifications. The 2.5G service transport delay in our case included TDM to Ethernet conversion delay at the customer premises, the EPON "serialization" delay, the core network transport delay, then any induced delay for Ethernet to TDM conversion (buffer delay jitter) for the TDM frames prior to transmission out the interface at the other side of the customer connection.
With many providers requiring 8 ms one-way delay, out of which 2 ms (worst case) was consumed by the TDM-to-Ethernet conversion process, 1.5 ms (again, approximate worst case) for EPON upstream timeslot transmission timing (approximately 500 ns or 0.5 ms), approximately 4.5 ms remained for end-to-end metro transit. This was just enough, though operationally we had to ensure the end-to-end delay metrics between the WSC and cell-tower were within 4 ms one-way delay.
Performance MTU size
Jumbo Ethernet frames are required for two reasons. The first is the requirement to deploy an MPLS-capable outer frame (the transport) that is expected to ingress minimum 1,526-byte frame into its Ethernet interface. MPLS-pseudo-wire emulation (PWE) encapsulates the ingress Ethernet 1,526-byte frame into it and forwards it through the ONU upstream. The minimum frame size requirements are in that case approximately 1,548 bytes.
The second is that many (perhaps all) of the cell providers specify a requirement for 1,600-byte frame support with plans for providing 9,000+ byte frame transport. To date, all providers have been willing to forego the larger frame size requirements (as long as we can support 1,526 or 1,530) because of the more reasonably priced EPON offering. In the future, we expect that some wireless providers will be providing their own PWE encapsulation, which will bring the supported MTU requirements up again to 1,570. Currently, we have MTU transport support of 1,600-byte frame size across our ONUs/OLTs, so we are not currently anticipating any inability to meet the wireless providers’ ultimate requirements.
The apparent reason for the larger 9,000-byte frames is efficiency. But as with any deterministic network (be it T-1, ATM, or in this case EPON), there must be a maximum frame size in order to budget and schedule fixed worst-case delays. This is the case for EPON and all of the PON technologies to date. So-called "jumbo" frames are not a standard and cannot be designed around. In this case, the requests for 9,000-byte frames appear to be requests for efficiency rather than a technical requirement.
EPON is the ideal technology for 3G and especially for 4G backhaul transport. The tree architecture of a PON is the best network architecture for the high density of cells that are required for 4G technologies. All 4G technologies will have a higher density of cells to support higher numbers of subscribers and greater data rates. All of the 4G technologies also feature support for Ethernet as the primary base station network interface. EPON is the only PON technology that supports transparent Ethernet transport compliant with the Metro Ethernet Forum standards. It can transport the access from one or a few users on a femtocell within a residential FTTH installation or hundreds of megabits per second to more than a 1Gbps for multiple-tenant cell sites. As much as Ethernet and EPON are replacing T-1s in the enterprise today, so too will they replace T-1s in cell backhaul.
Victor Blake is an independent consultant. Reach him at [email protected]. Edwin Mallette works for Bright House Networks. Reach him at [email protected].