August 2001
 

Bringing Ethernet And Optics Together

BY TONY RYBCZYNSKI

RINGING IN THE FUTURE
Various implementations of RPR can be envisaged, including integration into routers, Ethernet switches, and optical platforms. In the latter case, RPR is envisaged as a distributed Layer 2 switching architecture that provides reliable packet transport over an optical ring topology. The bandwidth on the ring is highly scalable, and configurable to run over a 10 Gb fiber or SONET physical layer (or potentially over lambdas in the future). The users could connect to RPR rings over a standard Ethernet User Network Interface (Ethernet UNI) operating at 10/100/1000 Mbps. These interfaces support IEEE802.1Q/p which enterprises use to support VLAN and QoS capabilities, respectively. "Users" could be individual PCs or servers, but more likely would be aggregation devices such as Layer 2 switches, Layer 3 routing switches, or conventional routers. They could be locally attached, as would be the case if the RPR looped through the customer building, or remotely attached using Ethernet over dark fiber.

Each multiport switched blade sitting on an RPR is a node. Each node auto-discovers other nodes on the ring through a topology discovery protocol that is initiated anytime the ring node map changes. The gathered information is used to create a database indicating the number of hops to each node in either direction. Each node can therefore determine whether transmitting a packet in a clockwise or counter-clockwise direction is best to reach another node on the ring over the least number of hops. Load sharing can be configured to allow bursts at twice the line rate (e.g., 1 Gbps over an OC12/622 Mbps ring) by transmitting data in both directions simultaneously.

RPR operates in a statistical multiplexing mode, allowing multiple traffic flows to share the ring bandwidth. When transmitting packets onto the ring, a fairness protocol is used to ensure equitable treatment across traffic accessing the ring. The definition of "fairness" is being actively debated. In addition to congestion management, other applications of fairness include congestion avoidance, Service Level Agreement support, bandwidth management, and delay/jitter optimization. In addition, multiple hardware queues can be implemented to support IEEE802.1p priority levels. If congestion occurs, simple on/off flow control, for example using IEEE802.3x to the user ports, can be applied, this being part of the Ethernet standard.

Every packet transmitted on an RPR ring contains a MAC address of the destination device. At each node, the MAC address is examined and if it has reached its destination, the packet is taken off the ring thus preserving ring bandwidth. This spatial reuse differentiates RPR from other ring architectures such as token ring or FDDI in that bandwidth is consumed only on traversed segments. With RPR, multiple nodes on other parts of the ring can transmit concurrently and hence create statistical gain for increased bandwidth utilization. RPR technology is very effective in handling multicast or broadcast packets, in which case a copy of the packet is taken off at each node. Unicast and multicast messages not destined for a particular node, and also broadcast packets, can be cut-through switched across the node. Such cut-through switching reduces transit delays around the ring to microseconds and virtually eliminates variations (jitter), making RPR particularly friendly to real-time IP traffic.

Another key attribute of RPR is fast rerouting in case of switch or link failures. Each direction of an RPR carries working traffic; there is no dedicated bandwidth put aside for protection. Fault detection is done through the use of special signaling packets sent periodically between neighbors in both directions on the ring. Lack of receipt of consecutive messages results in a declaration of link failure, with packets sent the other way around the ring. Continuous fault detection ensures fast detection, protection, and restoration (i.e., in less than 50 msec) providing a seamless service to Layer 3 devices.

RINGING FOR SUCCESS
For enterprises, running Ethernet-originated traffic over RPR combined with Ethernet on dark fiber and on DWDM can provide highly robust Optical Ethernets across the MAN and the WAN. This is one of the prime applications of RPR. In fact, this is already being done by enterprises ranging from a municipality in Virginia to a Toronto-based financial service institution. Running RPR over SONET pipes allows enterprises to establish Optical Ethernets that span the continent. Nortel Networks' corporate network is carried on four Resilient Packet Rings among its major North American sites. The immediate benefits to these enterprises are increased speed, reduced latency, and dramatic simplification of router configurations and engineering. Wide deployment of Optical Ethernets across the enterprise can impact the enterprise as a whole, by allowing them to rethink their deployment of servers, storage, and routing. This, in turn, can translate into more effective use of IT resources and increased agility in delivering new e-business applications.

Outsourcing is a growing trend in the industry, whether at the networking, application, or storage levels. Managed Virtual Private Ethernet services are Layer 2 services which provide low latency, protocol independent connectivity among enterprise sites. These services use RPR as a primary transport technology, Ethernet on fiber for access to customer sites and DWDM for bandwidth scalability. Secure segregation of traffic from different customers is achieved through the use of standard-based encapsulation carrying a unique VPN customer service identifier. In addition, Virtual Private Ethernet implementations define mechanisms to deliver to enterprise service level agreements, which incorporate end-to-end reliability, committed access rates, and latency guarantees. Committed access rate applies to an entire Ethernet interface and can be configured for speeds from 1 to 1,000 Mbps in 1 Mbps slices, while supporting wire speed bursting. Dynamic configuration of these access rates can be provided to customers through a simple Web interface. Scalability to tens of thousands of customers and millions of ports is achieved by selectively exploiting various IP networking capabilities, ultimately leveraging MutiProtocol Label Switching (MPLS) to provide scalable, standards-based, Virtual Private Ethernet solutions.

Resilient Packet Rings free Ethernet traffic from the bonds of distance limitations over point-to-point links, and expands it to run over robust optical ring architectures ultimately spanning the globe. It is a key technology that enables enterprise Optical Ethernets built by enterprises or procured as managed Virtual Private Ethernet services. In either case, enterprise Optical Ethernets not only simplify the networking environment but open up opportunities for routing, storage and server centralization, low latency network computing, and improved overall knowledge worker productivity.

Tony Rybczynski is director of strategic marketing and technologies for Nortel Networks' Enterprise Solutions unit. E-mail questions or comments to tonyryb@nortelnetworks.com.

[ Return To The August 2001 Table Of Contents ]