Delivering fiber connections to the home or curb has been a hot subject in the communications sector over the past few months. With the price of fiber decreasing and DSL and cable modems still struggling to reach customers, many are pointing to passive optical networks (PONs) as the ideal solution for bridging the often-talked-about gap between the central office (CO) and end users. In fact, a host of carriers has started running fiber-to-the-home/fiber-to-the-curb (FTTH/FTTC) trials to evaluate the viability of using PONs as a means to deliver broadband services. But for FTTH/FTTC networks to take off, the design community first must resolve a key technical debate. Right now, both ATM-based PON (APON) and Ethernet-based PON (EPON) technologies are being pushed as the answer for delivering FTTH functionality. Each of these technologies, however, has its own advantages and drawbacks.
So, what's a PON?
Before going any further, though, what is a PON? Briefly, a PON consists of an optical line terminator (OLT) located at the CO and a set of associated optical network terminals (ONTs) located on the customer's premise. Between them lies the optical distribution network (ODN) comprising fibers and passive splitters or couplers (Figure 1).
In the PON structure, one OLT can have multiple PON units, each of which drives a single-mode fiber through inexpensive passive optical splitters and distribution fibers to connect to multiple ONTs. The fibers and passive optical splitters eliminate the need and the associated maintenance cost for active electronics in the distribution facility of the access network.
The process of transporting data downstream is different from transporting data upstream. Data is broadcast downstream, and each ONT takes in a packet destined for it by matching the address at the packet header.
Upstream traffic in an APON is transmitted at the control of the OLT using a TDMA (time division, multiple access) protocol, in which transmission time slots are granted to individual ONTs. The time slots are synchronized so that packets traveling upstream from the ONTs do not collide with each other when hopping onto the common fiber.
At the request of the OLT, an ONT may also report to the OLT its packet-buffer length, and the OLT can subsequently reapportion grants accordingly to accommodate the traffic needs of the ONT. Wavelength-division multiplexing (WDM) is used.1
The APON (ITU-T G.983.x) was the first PON implementation. To ensure consistent PON services and bandwidth control, the OLT and its associated ONTs are managed as a single network element. As an access technology, APON is transparent to user applications and is flexible at providing both packet- and circuit-based services for a complete range of all currently known and new services for business and residential customers.2
The EPON proposal is being studied within the IEEE 802.3 working group, through the "Ethernet in the First Mile" task force. It's important to note that the Ethernet in the First Mile project includes three approaches; (1) point-to-point copper over the existing copper plant at speeds of at least 10 Mbit/s up to at least 750 m, (2) point-to-point fiber over a single fiber at a speed of 1 Gbit/s up to at least 10 km and (3) point-to-multipoint fiber using PON at a speed of 1 Gbit/s up to at least 10 km. The third approach is referred to as EPON.
The fundamental differences between APON and EPON stem from their packet transmission approaches.
ATM overhead tax
In APON, data is transported via fixed-size packets of 53 octets. For APON to carry any traffic, data must go through an ATM adaptation layer (AAL), where it is segmented into a 48-octet unit with a five-octet header added, resulting in 53-octet packets called an ATM cell. When the cells arrive at the destination, they are then reassembled back into the original traffic. This is called the segmentation and reassembly (SAR) process.
On the one hand, SAR allows ATM to be easily adapted for voice, video and data services. Its small, fixed-size packets are deemed essential for carrying delay-sensitive traffic.
On the other hand, the SAR process can be time-consuming even if implemented in hardware. The provisioning of ATM virtual paths and virtual channels and associated cross-connections also appears to add further complexity to the ATM protocol. The five-octet header for a 48-octet payload is often called the ATM cell tax, of about 10 percent.
While some have criticized ATM for its complexity, it turns out that these fixed-size packets are a blessing for the PON upstream TDMA mechanism. Because the packets are of one fixed size, the upstream traffic control is simplified, allowing the OLT to grant upstream time slots to individual ONTs so they can send data upstream on the common fiber without collision. As the distances between the OLT and ONTs vary, individual ONTs are configured to insert appropriate delays whose values are derived by a process called ranging.1
APON networks ensure collision-free upstream transmission by requiring all ONTs to transmit cells at the equivalent maximum distance of 20 km. Specifically, to discover the equalization delay for each ONT, the OLT sends a ranging grant to an ONT and waits for a response.. The round-trip time delay is measured and subtracted from the maximum 40-km round trip time delay, producing the equalization delay for the ONT. This method is straightforward and effective because of the fixed packet size.
The EPON technique
In contrast to APON, EPON data is transported in variable-length packets from 64 up to 1,518 octets based on the IEEE 802.3 Ethernet frame format. Because of variable-size Ethernet frames and the traditional carrier sense multiple access/collision detection (CSMA/CD) Ethernet protocol, the upstream EPON TDMA protocol would be very complex. Thus, designers must strike a balance between TDMA and variable-size Ethernet frames.
One approach is to assign a maximum time slot that would fit all packets of all sizes. This clearly would cause concern for bandwidth inefficiency, even at gigabit rates. For instance, at the worst, putting 64-octet packets in the time slots for 1,518-octet packets would lead to an efficiency of about only 4.2 percent.
Another approach is to select a proper time-slot size and, when possible, let the ONT concatenate multiple Ethernet frames into one time slot. While efficiency will improve in this situation, there can still be bandwidth waste because it is not probable that variable-size frames will fit evenly in one slot. Concatenation could also lead to protocol complexity.
Under a third approach, designers can segment Ethernet frames into fixed-size packets for simplifying upstream traffic. One advantage is that there is much work available for reference. The price to pay would be adding a SAR layer onto the EPON protocol stack.
One other challenge with EPON packet transport lies in the Ethernet standard itself. The traditional Ethernet peer-to-peer relationship between network nodes does not exist among ONTs. The media access control (MAC) and physical layers (PHYs) of the existing IEEE 802.3 CSMA/CD Ethernet standard must be enhanced to adapt to the PON infrastructure and its associated traffic-flow paradigm of downstream broadcasting and upstream TDMA.
Data security and QoS
While broadband services deliver broadband pipes to the home, they also open end users' traffic to potential hackers. As with data transport, security is handled differently by EPONs and APONs.
Under the APON method, downstream traffic is churned point-to-point at the transmission-convergence layer to avoid complex encryption at the PHY. Each churning key is set and updated regularly by individual ONTs.1 In addition, the OLT may request passwords from a particular ONT to prevent access by a malicious user masquerading as another ONT. This security mechanism is part of the APON protocol.
The Ethernet protocol, based on CSMA/CD, does not have any built-in security mechanism. It's expected, however, that EPON developers will invent specific security methods to ease concerns about carrying traffic using the point-to-multipoint method.
In addition, EPON equipment developers can tap into multilayered security mechanisms for IP traffic, such as firewalls, VPNs, Internet protocol security (IPSec) and tunneling. The associated overhead cost with those methods may be trivial, especially as EPON systems begin pushing gigabit rates.
In addition to security, guaranteed QoS and dynamic bandwidth assignment (DBA) capabilities must be considered when evaluating APON and EPON. DBA significantly increases the throughput of the PON traffic. The downstream broadcasting of individual packets from OLT to the ONTs is logically equivalent to point-to-point transmission. Therefore, it does not present an issue for bandwidth control.
However, the upstream transmission of packets from multiple ONTs onto a common fiber requires an adequate bandwidth-control mechanism. APON is designed to accommodate combined bursty data and QoS-guaranteed traffic.
APON's DBA protocol involves two possible approaches. First, the OLT can perform idle-cell monitoring to gauge the grant use and adjust grants accordingly. Second, an ONT can indicate its bandwidth needs to the OLT, with the OLT adjusting grant assignment to meet those needs in real-time.
A hybrid method combining the two is also recommended. However, for what it does, the proposed APON bandwidth allocation protocol is fairly complex. It is the new ITU-T G.983.4 Recommendation.
The traditional CSMA/CD Ethernet protocol does not have any specific bandwidth management function, but many proposals for upstream traffic control have touched on the model of request/grant as adopted in APON.3
At this point, the guaranteed QoS has yet to be identified as one of the objectives for the Ethernet in the First Mile Task Force. Some researchers suggest that, at gigabit rates, EPON will rely on an IP traffic-management mechanism to offer the bandwidth and QoS capabilities as provided by APON.4 Some approaches include the use of differentiated services (DiffServ) and guaranteed QoS using a type-of-service (TOS) field specified by IEEE 802.1p, which identifies eight prioritized service levels. Those methods can be very effective for IP-layer traffic management.
But the methods do not adequately address the PON QoS concerns. In the EPON setup, QoS implemented at the higher layers is dependent upon Layer 2 for packet transmission. The upper layers, however, have no knowledge of the packet sizes and time-slot length at the PON layer.
Thus, an adequate bandwidth-control function for upstream traffic is still required in EPON systems to control delay, jitter and packet loss.
Figure 2 compares EPON with APON.
The APON effort is reaching a point of maturity, while EPON is in the early phases of development. EPON, however, faces interesting challenges, the most noticeable of which is to handle the upstream traffic, moving away from the traditional, collision-based CSMA/CD Ethernet protocol.
Author's Note: Thanks to Rick Heil, Mike Figueras, Yazid Benkhellat, Jeremy Smith, Hernando Valencia, Allen Benson and Eiichi Kabaya for their help on this article.
Zheng-Yang Liu is a principal engineer at NEC eLUMINANT Technologies Inc. He has a BS from Hefei Polytechnic University, China; an MS in computer science from the University of Maryland, College Park; and a PhD in computer science from the University of Oregon. He can be reached at liuz@eluminant.com.
References
- ITU-T Recommendation G.983.1, "Broadband optical access systems based on passive optical networks (PONs)," 1998.
- International Engineering Consortium Web ProForums, "ATM Passive Optical Networks (PONs)," www.iec.org/online/tutorials/atm_pon/, 2001.
- IEEE 802.3 EFM Internet site. www.ieee802.org/3/efm, 2001.
- Pesavento, G., and Kelsey, M., "Ethernet in the first mile-Access Solutions/Special Reports," Lightwave, June 2001.
