Resilient Packet Ring technology provides a cost-effective means of adding Ethernet services to SONET-based networks. It is also recommended as a standalone approach.
Gady Rosenfeld
The RPR Alliance
Corrigent Systems
Ethernet has become the de facto standard for data networking in the LAN. The vast deployments of Ethernet equipment in the LAN have made Ethernet the cheapest technology, on a dollar-per-megabit basis, to the end-user. This, in turn, has created a constantly growing demand for Ethernet services in the service provider network. Resilient Packet Ring (RPR), a new media access control (MAC) protocol being standardized in the IEEE 802.17 working group, is designed to deliver Ethernet services over the public network in the most cost-effective way. This business case analysis, which examines RPR over an existing SONET network as well as RPR as a standalone network for Ethernet services, is especially critical now given the economic conditions facing the telecom industry.
Using RPR for Ethernet over existing SONET capacity
Demand for Ethernet services is indeed growing, and established carriers are struggling to determine the most cost-effective architecture to deliver these services. In many cases, however, the demand for Ethernet remains relatively low compared with the demand for private-line services and doesn't justify the expense of building a separate Ethernet transport network. Moreover, established carriers have built high-capacity SONET networks in the past couple of years, and a significant part of this capacity is still unused today. For these reasons, some carriers wish to deliver Ethernet services over their existing SONET infrastructure.
The following demand scenario represents modest initial demand for Ethernet services. The scenario consists of 2xGigabit Ethernet (GigE) services per node, on a four-node metro ring, where each service is required to peak to 600 Mbits/sec with an average utilization rate of 100 Mbits/sec. All of the traffic in this case is headed outside of the metro ring (through a hub node). To analyze the sensitivity of the results to the specific demand scenario, this analysis presents results for a scaled-up demand scenario by a factor of 2 (4xGigE ports per node).
Several network architecture alternatives can be considered:
Direct Mapping of Ethernet over SONET: In this network architecture, end-users connect directly to Ethernet ports on an Ethernet blade in a SONET add/drop multiplexer (ADM). No aggregation is performed at the edges; at the hub, Ethernet traffic is handed over to an Ethernet switch for aggregation and switching. In this scenario, two STS-12 circuits are provisioned from each node to the hub to support two Ethernet services (bursting each to 600 Mbits/sec) per node.
Ethernet Aggregation at the Edge: In this network architecture, end-users are connected to a low-cost, enterprise-class Ethernet switch that aggregates user ports and hands over traffic to a SONET ADM. Ethernet traffic is carried over SONET to the hub for further aggregation and switching. In this scenario, two Ethernet services are aggregated to one Ethernet port using the low-cost Ethernet switch, and an STS-12 circuit is provisioned between each node to the hub.
RPR Distributed Aggregation over SONET: In this network architecture, end-users are connected to an RPR ADM that aggregates user ports from all over the ring over a single shared ring. At the hub, the RPR ADM just drops the Ethernet traffic heading out and hands it over to a long-haul network. In this scenario, an RPR ADM is deployed over a chain of virtually concatenated circuits that create a single shared RPR ring. The size of the shared ring can be configured according to the total amount of committed traffic.
Figure 1 demonstrates the differences between each of the discussed architectures, and Figure 2 presents the required SONET capacity for each of the architectures and for the two demand scenarios.
Economic analysis
This business case analysis considers the required additional equipment on top of the existing SONET equipment as the basis for the assessment of the expected capital expenditure. The quantitative analysis presented above demonstrates why employing RPR is the only practical way to use existing SONET capacity for the transport of Ethernet services, and any other option involves deployment of additional SONET equipment (which obviously translates into additional capital expenditure).
- Direct mapping of bursty Ethernet traffic will consume the available SONET capacity immediately after provisioning the first Ethernet service.
- Local aggregation of Ethernet services (at the node level) will improve the overall utilization but will still involve deployment of additional SONET equipment, even for an existing OC-48 SONET ring that is only 25% utilized.
- Employment of RPR enables to use the existing capacity without the need to deploy additional SONET equipment.
Unlike private-line services, Ethernet services are bursty in nature and are expected to run at a significantly lower tariff (per Mbit/sec) than TDM-based private line services. Hence, the only way service providers could gain an increase in its revenue stream when Ethernet services are deployed is if they can efficiently aggregate and transport the bursty traffic over their transport infrastructure.
Obviously, mapping bursty Ethernet traffic directly over point-to-point SONET circuits will not enable service providers to offer Ethernet services at lower tariffs than private-line services, if they want to keep their revenue streams flat. On the other hand, the efficient aggregation an RPR transport layer provides can enable a service provider to actually increase its revenue stream (per Mbits/sec) by deploying Ethernet services.
Using RPR as a standalone solution
As demand for Ethernet services grows, a service provider may choose to dedicate a separate network to transport these services. To demonstrate the scalability of each of the network architecture alternatives, the analysis below uses a rather aggressive demand scenario that consists of 16xGigE services per node on an eight-node metro ring, where each service is required to peak to 1 Gbit/sec with an average utilization rate of 100 Mbits/sec. Most of the traffic (80%) is headed outside of the metro ring, which makes this scenario representative of an Internet access-based service offering.
Here, the network architecture alternatives are different than those discussed before:
Ethernet over SONET: In an Ethernet-over-SONET architecture, Ethernet traffic is aggregated and switched at the node as well as at the hub using Ethernet switches, while it is transported across the metro over traditional SONET infrastructure. In each node an Ethernet switch aggregates bursty Ethernet traffic into GbE trunks that are handed over to a SONET ADM, mapped over circuits, and carried over SONET to the hub. At the hub, Ethernet flows are recovered and handed over to a high-capacity Ethernet switch that performs both aggregation into uplink trunks for inter-metro traffic and switching for intra-metro traffic.
Ethernet Hub-and-Spoke: In a pure Ethernet-based transport architecture, Ethernet switches aggregate traffic at the nodes and switch it to long-reach trunks that are all connected to a high-capacity Ethernet switch at a central hub.
RPR-Based Distributed Switch: In an RPR-based distributed switch architecture, aggregation is performed locally -- both on the node's traffic as well as on the shared media ring. A hub node is required to backhaul inter-metro traffic to the backbone network but not for intra-metro switching of traffic.
Figure 3 illustrates the three network architectures, and Table 1 (below) presents the main cost components that influence capital expenditure for each of those alternatives
Table 1: Main Cost Factors for the Three Alternatives| Ethernet over SONET | Ethernet p2p | RPR | |
|---|---|---|---|
| Fiber Pairs | 8 | 8 | 1 |
| Trunk Optical Ports | 32xOC-192 | 32x10GbE | 18xRPR/OC-192c |
| Trunk Reach | LR | LR | SR-2 |
| GbE Ports | 320 | 128 | 128 |
Analysis
An RPR-based metro transport network delivers SONET-class five-nines availability for all services running over it for half the price of a comparable point-to-point Ethernet network. The reason is the fundamental architectural difference between an RPR network and a point-to-point -based architecture, resulting in an inherently different equipment configuration, which greatly influences the total capital cost required.
A network based on a ring topology uses half the number of optical ports than does a comparable hub-and-spoke topology. A metro network that consists of N nodes and a hub and carries resilient services would require 2(N+1) optical ports when it is deployed in ring topology, and 4N ports when it is deployed in a hub-and-spoke configuration.
A network based on a ring topology also uses shorter-reach optical ports. In a ring topology, the optical signal is regenerated in every node, whereas a hub-and-spoke logical topology over a physical fiber ring requires long-reach optical ports. Optical ports are the most significant cost factor in today's high-rate transport equipment and typically account for half the equipment cost.
Since fiber is deployed in a physical ring topology, a logical hub-and-spoke network uses N times the fiber mileage, or number of wavelengths, than a comparable logical ring network.
In a point-to-point network, all traffic is switched through a centralized hub that has to support peak burst rates of all the flows carried over the metro network. In contrast, RPR's bandwidth management scheme functions as a distributed switch and obviates the need for additional switching capacity for intra-metro traffic.
Finally, the bandwidth management scheme RPR offers enables efficient statistical multiplexing over the total amount of traffic carried in the metro at the local node rather than in a distant location, thus greatly alleviating the costs of backhauling traffic. A comparable network that lacks RPR's bandwidth management scheme would be able at best to employ a much less effective statistical multiplexing on the node level, resulting in more traffic that has to be backhauled to a centralized hub for further aggregation.
Summary
An RPR-based transport network can save significant capital expenditure compared with other alternatives in the market, both when deployed over an existing underutilized SONET network and as a standalone transport solution.
Compared with alternative solutions, an RPR-enabled transport architecture efficiently aggregates bursty Ethernet services over the shared media ring, resulting in a significantly lower amount of SONET resources required to carry bursty Ethernet services. The difference in the required SONET capacity translates into significant savings of capital expenditure for an RPR-enabled overlay approach compared with an alternative Ethernet point-to-point architecture.
No less important is the effect an RPR-enabled architecture has on the expected service provider's revenue stream from its transport network. SONET capacity that is used to carry Ethernet services cannot be used to carry high-margin private-line services - a dramatic reduction in revenue potential per Gbit/sec deployed. To make the Ethernet business case possible for carriers, Ethernet services must be efficiently carried over the SONET infrastructure such that the dollar-per-Gbit/sec consumed is positive relative to private-line services.
Gady Rosenfeld is Director-Strategic Marketing at Corrigent Systems (San Jose, CA). He can be reached at [email protected]. More information on the RPR Alliance can be found at www.rpralliance.org.
