Metro networks: the future mode of operation
Oct. 16, 2001--EXCLUSIVE--How are carriers responding to the evolving changes in the metro and long haul? Service providers need to know how to approach these challenges to get the most out of network performance.
By Hagay Katz co-founder and Director of Product Management, PacketLight Networks Ltd.; and Dr. Michael Mesh, co-founder and CTO, PacketLight Networks Ltd.
How are carriers responding to the evolving changes in the metro and long haul? Service providers need to know how to approach these challenges to get the most out of network performance.
A "Lambda Storm" in the Long-Haul
Although "winds of change" are natural in telecommunications networks, recent developments in long-haul networks can be seen more as a high storm. The typical models of SONET capacity upgrade from OC-12 to OC-48 to OC-192, and the typical DWDM capacity upgrade from one lambda to many, have changed. A new model has come into place where a whole lambda (typically OC-48) is seen as a network resource and can be provisioned on an end-to-end basis using very fast signaling techniques. Moreover, protection of these lambda resources is done on a network basis, instead of using a string of point-to-point and ring segments. As the majority of these wavelengths carry data services, which do not require stringent SONET protection, it is possible to provide protection based on alternative restoration schemes that better utilize lambda resources.
"Ethernet Storm" in the Enterprise
While thousands of OC-48 pipes have hit the long-haul segment, the enterprise market has been caught in a different storm. Ethernet technology has made bandwidth a commodity. "Ridiculously fast -- ridiculously cheap" GigE technology has pushed aside ATM, changing the rules of the game. Layer 2-3 bandwidth managers enable provisioning of bandwidth to customers in any granularity. WAN interfaces in the routers, which are typically POS-based or ATM/FR, are expensive and customers want to use the Ethernet interfaces as their preferred method to connect to the WAN. It is now estimated that, in the next five years, Ethernet will comprise up to 40 percent of the traffic from enterprise networks towards the metro transport.
The Unpredictable Metro
The advent of these storms has caused metro network designers to be faced with a period of uncertainty.
(a) Uncertainty in the metro access: It is clear that GigE packet data is coming, but it is not clear when and from which location. TDM based voice is still the most reliable profitable and resilient service offering and will continue to be needed by customers. ATM is still a very important technology due to the investments made in DSL. Forecasting the type of traffic and interface cards coming from customer sites has become an issue. Service providers that will be able to provision any service with minimal installation and network provisioning changes will win the customers.
(b) Uncertainty in the metro core: GigE core routers with new interface cards such as GigE and POS are taking the place of BB cross-connects and ATM switches. But at what rate? And where exactly in the network? and is it limited to data traffic only?
The traditional way to cope with this uncertainty is to provide smart devices that implement layer 3-7 processing at the edge of the network. These devices sort and terminate the different traffic types close to the customer site and channel the specific traffic over dedicated lambdas to their associated core switches and routers. The big problem with these devices is that they fight the uncertainty at the edge, where the uniformity of demands is lowest. Therefore, every edge must be prepared for the worst-case scenario. Even if the proportions between GigE, ATM, and TDM traffic are unknown, STS Cross-connects, IP and ATM fabrics are installed "waiting to terminate" the "enemy" cells, packets and time-slots.
Of course, this strategy has its rationalization. Until now, there was no solution available to handle GigE traffic as well as ATM and TDM on one switch fabric and on the same lambdas. Conversion from GigE to ATM does not make a lot of sense due to costs and capacity issues. Additionally, SONET STS cross-connects cannot handle packets efficiently. Next generation SONET devices built a few layers on top of the SONET cross-connect to map GigE and ATM to SONET. But this mapping is not efficient, and the transport networks do not make efficient use of the fiber. More importantly, provisioning remains very difficult. The SONET rigid allocation of time slots requires node-by-node planning and changes when a new service is implemented. Conversion to an all-IP network seems like a good solution, theoretically, but conversion to Layer 3 and back is also very painful and costly.
The logical way to cope with this uncertainty is to have a Layer 1-2 transparent unified service collector at the edge and a single point of aggregation at the core.
Metro: Future Mode of Operation
Layer 1-2 unified service collector: A new method, which allows native form unified fabric, is required. It involves no conversions and uses a single fabric at the edge. By performing layer 1-2 unified packet adaptation, this method keeps Ethernet, ATM, and TDM in their native mode without terminating them and sending them to the network core. This means that any service can be collected by simply adding a new adaptation module. This kind of all-packet metro edge can be connected to other nodes in a packet ring or a point-to-point link. There is no need for costly multi-service fabric systems and multi-layer next generation SONET.
Single point of aggregation: Metro core and metro access networks are built in such a way that many POPs are directed to one or two hubs. Since the uniformity at the core is higher than at the edge, the law of big numbers dictates that uncertainty at one edge will be offset by uncertainty at another edge. Moreover, reducing the complexities at the edge is paramount to achieving network management simplicity.
Optical Networking in the Metro
Each node performs Layer 1-2 packet switching, based on the MPLS. The nodes are connected in packet rings with SONET (or other) restoration schemes. Services are collected and forwarded to the aggregation point. Using MPLS, each packet is assigned a different QoS and has an end-to-end route based on the provisioning of the service. Using this kind of all-packet technology meets the two main challenges faced by service providers:
Simple provisioning: Provisioning a service on an end-to-end basis is as simple as setting up an end-to-end MPLS connection. There is no need to rearrange rigid SONET allocations, as there are only fat packet pipes along the way from the access to core. If there is enough capacity, provisioning is done by simply updating the forwarding tables.
Optimal fiber utilization: Services can be packed into wavelengths using any combination (e.g. OC-12, Fiber Channel and GigE on the same lambda) and at any point in the network.
In such a network each node is an LSR and MPLS labels are distributed using automatic CR-LDP protocols or centralized network management functionality.
A new mode of operation and a new class of systems is required to allow carriers to better respond to the changes in the enterprise and long haul networks. All-packet transparent metro nodes that employ packet and optical networking, allow carriers to achieve cost effective, simple to manage, efficient transport of multiple services on their fiber infrastructure.
About PacketLight Networks:
For more information, visit www.packetlight.com.