But what Matters Is Their Impact on IP-based Services - Technology Information

Telecommunications, April, 2001 by Cecil Christie

Demand for bandwith-intensive applications such as dynamic, streaming content from residential and small business customers continues to escalate, forcing service providers to seek new ways to keep up. One solution is to deliver fiber access, which offers the promise of bandwidth abundance. PONs are a step in the right direction.

However, PONs only deliver the optical transport. Additional technologies must be overlaid on PONs for full-service delivery. While the technology choice of ATM or Ethernet as a transport over PONs is important, the bigger question is the impact of that choice on IP-based services running on the network.

The APON Service Model

The issues for APON center around the ability of an end-to-end channel to deliver strict service and security guarantees as it traverses the shared PON medium. APONs also suffer from framing overhead that further reduces bandwidth in a shared environment. Bandwidth guarantees may be difficult to meet given the shared nature of the PON. Adding new customers compromises guarantees given to existing customer applications, if the PON bandwith is already fully subscribed. Provisioning and reprovisioning the service may add administrative overhead. Could these additional costs neutralize the cost benefits gained from using a shared optical medium?

The EPON Service Model

In the past, one big drawback to Ethernet deployment in shared environments was its inability to classify traffic. Recently, however, the IEEE delivered on the 802.1p standard to address prioritization issues. Additional enhancements to Ethernet may make it possible to provide virtual channels for customer traffic and also perform OAM&P functions.

Another approach to creating channels is to couple WDM, Ethernet and IP. Each channel then serves as traffic segregator and carries Ethernet frames. Assigning IP ToS (types-of-service) bits per ingress wavelength and Ethernet packet would then be straightforward.

IP Services Over APON and EPON

The proliferation of the Internet is definite proof that no transport-layer technology can be provisioned end-toend. Service guarantees are best offered and delivered at the layer that spans networks to the end points. Since this layer is IP, the issue is how well IP works over APONs or EPONs.

Services that interest enterprises, such as data storage and backup, intranets, extrants and content and application delivery, require a broadband infrastructure. These services cannot be provisioned and managed at the link layers. Previous-generation services, which supported the "charge-by-the-bit" business model, are not relevant for enterprise services. The layer for content-based services must provide security, management, monitoring and accounting at the IP layer. The services that can be deployed and managed at this layer are affected by the choice of APON or EPON.

Providers delivering services at the LP layer with ATM and APON links could choose to implement them through MPLS (multiprotocol label switching) VPNs. In this case, bandwidth allocation and other provisioning occurs at the IP layer, reducing the need for those functions at the transport layer. In fact, for MPLS implementations, ATM switches in the core also function as label switch routers for IP. For non-MPLS-based IP implementations over ATM, traffic management functions could be handled by ATM with mappings between IP ToS and ATM traffic type occurring at the boundary of the ATM network. Such networks need careful design to ensure that service quality is maintained during the handoff. Thus, IP networking over APON would require maintaining and managing two overlaid networks. Since ATM is also functioning as a full-service layer, the ratio of overhead to actual data tends to grow large without offering significant benefit.

By comparison, in IP over EPONs, IP is the sole service layer. This makes it possible to deploy a uniform service model throughout the network. This also homogenizes packet delivery as traffic originates as IP over Ethernet and gets transported as IP over Ethernet. A hosting provider can structure a business around Web site uptime guarantees, Web page load times, and streaming video latencies assuming a homogenous service network. Storage service providers can also offer per-transaction guarantees and bulk-transfer services independent of network-level translations.

Ironically, if PONs prove to be too easy to deploy, they might become a victim of their own success. As consumer demand for bandwith continues to rise, PONs and their passive components will have to incorporate more and more intelligence to manage the higher traffic loads. This might require adding multiple fiber pairs for resiliency and switching technology within the splitters. By taking these measures, service providers will be able to offer greater bandwidth to each user and deliver cost-effective service creation and management. The very nature of PONs will evolve as it takes on characteristics of intelligent, services-aware networks. However, for many years yet, PONs hold great potential as the answer to a question perplexing service providers for years: how to carry fiber and its wealth of capacity across the last mile.