InFocus: How smart is your network?

While traffic demands may force you to assign bandwidth
in any way that will get the job done, giving the layers
of your network more information will help maximize efficiency

The phenomenal growth of data networks, particularly the Internet, is constrained only by the ability of service providers to deploy equipment.

RESOURCES From Our Archives

While this growth presents a major opportunity for service providers, capitalizing on it presents considerable challenge. The continuing cycle of new users, broadband access technologies and high-bandwidth applications is threatening to outstrip the capacity of current networks. Rapidly changing traffic patterns make it difficult for service providers to configure their networks efficiently. And new mission-critical applications need a higher level of dependability--one that is costly to offer using current network infrastructures.

To meet these challenges, service providers are re-evaluating how they construct and manage their networks. Many are moving towards a new, backbone network architecture that enables intelligent interworking between the packet and optical layers of their networks.

Today's Networks

Traditionally, networks have been built using a multi-layer model. An IP network, for example, may run over a frame relay network, that runs over an ATM network, which runs over a Sonet/SDH network, which runs over an optical (wavelength) network. The advantage of this approach is that the layers can evolve independently, while continuing to support legacy services. There are, however, disadvantages to this multi-layer solution. The large number of devices in today's networks increase network cost and complicates the development, deployment and management of new services.

Each layer of the network typically has an independent management structure and associated processes that only have visibility of the topology and state information of that one layer. The number of management systems increases network cost while adding complexity to network-wide operational tasks such as provisioning, performance monitoring and fault isolation. Two primary difficulties with the multilayer model are the inability of service providers to optimize the use of network resources and to provide cost-effective network reliability.

Network Optimization

A service provider's competitive advantage is, in large part, dependent upon its efficient use of network infrastructure. It is not surprising, therefore, that service providers spend significant time and resources optimizing their networks to increase revenues (through the addition of new services) and decrease costs (by making their networks more efficient).

Because the packet layer in today's networks is independent from its lower-level, optical counterparts, today's packet-network topology is limited to pre-established optical connections. Consequently, the packet network does not have the ability to use the optical layer to establish optical paths intelligently across the network. Thus, if new traffic demands appear, and cannot be provisioned along the most efficient route due to capacity restraints, the service provider has two options:

• engineer and provision additional capacity along the most efficient route (which may take weeks or even months), or
• Provide the service through a less optimal, alternate route that has spare capacity.

Unfortunately, the requirement to provision the service quickly often drives the service provider toward the second option, which results in a less efficient use of packet and optical resources.

Rapidly changing traffic patterns also create network inefficiencies. Many analyses have shown that the predictability of the Internet traffic is changing, for the worse, all the time. The prediction techniques that almost worked three years ago are now useless. As a result, service providers constantly monitor their traffic to identify opportunities for network optimization.

When changing traffic patterns require reconfiguring their networks, service providers must manually engineer and provision new connections at both the packet and optical layers of the network. Many large ISPs undertake this process on a weekly or daily basis during non-peak network usage.

Unfortunately, this manual network reconfiguration is inefficient because:

• the network's optimization lags the demands of the network traffic;
• it is operationally expensive;
• it requires significant numbers of network engineers; and
• It introduces the potential for human error.

Network Availability

Services providers face a tremendous challenge in building available and maintainable, next-generation, public-carrier networks. At the packet layer, IP has become the networking protocol of choice. Current IP networks, however, have yet to meet requirements for mission-critical services. An independent University of Michigan study indicates that router-based, wide-area, backbone networks are ten to 1000 times less available than the traditional public switched telephone network (PSTN).

If all of today's traffic will be carried over the IP infrastructure, it is clear that the availability of the IP networks must be improved significantly. Vendors in this space typically quote 5 9's (99.999%) availability. This statistic, however, often only refers to their hardware reliability. While some vendors also claim non-service affecting software upgrades and software rollbacks, network congestion and physical link failures (which together cause of 33% of network downtime, according to the University of Michigan study) are overlooked.

To solve network congestion problems, service providers must again engage in regular network reconfigurations on a manual basis. The problems with this process are mentioned above. To protect against physical link failures, service providers typically employ Sonet/SDH, which provides automatic protection switching (APS) capabilities. While APS does an excellent job of protecting against physical link failures (switching traffic to an alternative path is less than 50 milliseconds), it is resource intensive, requiring that large amounts of spare bandwidth be held in reserve in the event of a fault.

Initially, when Sonet's primary purpose was to protect voice and private line traffic, this level of resource investment was not an issue. The return on investment for these applications was worth it and applications that did not require this level of robustness were not prevalent. In today's network environment, however, there is more variation in applications...including many that do not require premium quality of service.

For this reason the optical layer is evolving to provide more flexible ring and mesh architectures. Ring architectures will enable full APS protection for high priority traffic while providing access to protection capacity for low-priority traffic. Mesh architectures are evolving to reduce the amount of protection capacity used in the network for low priority traffic.

Smart Packet/Optical Interworking

The key to successfully dealing with these network optimization and reliability concerns is the concept of "smart packet/optical interworking." Smart packet/optical interworking represents the sharing of information between the packet and optical layers of the network. Beyond merely sharing information, however, it enables the network to take action based on the shared information, and to optimize the network dynamically.

One approach to achieve smart packet/optical interworking is through the use of an intelligent "network controller" function. The network controller monitors network state and topology via connections to both the packet and optical network layers. Network state information comes from the packet layer in the form of link usage, loss, latency and congestion statistics. Network topology information comes from the optical layer in the form of wavelength usage, failures and alarm statistics.

Using the network state and topology information, in conjunction with pre-defined policies created by the service provider to maximize profits, the network controller would dynamically detect opportunities for network optimization. For example, if the network controller recognized that packet loss on a specific multiprotocol label switching (MPLS) path exceeded a given threshold, it would assess potential solutions and identify the optimal one. Under the circumstances described, the network controller might consider increasing the size of the MPLS path at the packet layer; rerouting some portion of the traffic onto a second MPLS path; or providing the MPLS path with additional optical resources, such wavelengths or synchronous transport signals (STSs).

Another function of the network controller is to report its recommendations to the service provider through a network management system. Service providers could then use this information to optimize their networks.

While service providers are likely to introduce the network controller function into their networks gradually, it will ultimately take on an additional role-automatic reconfiguration. This automatic reconfiguration would eliminate manual intervention on the part of the service. It could occur either at the packet or optical layer of the network. Most likely, the network controller would begin at the packet layer, where it would attempt to optimize the network through MPLS changes.

For example, it could add an MPLS path, change the characteristics of the path, or reroute an existing MPLS path. If changes at the packet layer are insufficient to optimize the network, the network controller has the option of moving to the optical layer where it could establish, break or move optical connections.

It is important that the network controller mediate the flow of information between the optical and packet layers without requiring direct coordination between the devices or common product design philosophies. Such an approach permits both the packet and optical layers to aggressively incorporate new technological innovations independently. This is critical as optical and packet layer technologies lie on very different performance improvement curves. In addition, separation between the packet and optical network elements prevents the packet-layer device, presumably a very large IP router or MPLS switch, from wasting cycles unnecessarily on tandem traffic.

Smart packet/optical interworking provides service providers with a number of important benefits. First, it automates network optimization. No longer are service providers required to employ rooms full of traffic engineers, increasing the potential for human error. Moreover, its dynamic nature enables service providers to decrease the time between changes in traffic patterns and network reconfiguration, thereby maximizing the use of network resources on an ongoing basis. Finally, because the network is dynamically reconfiguring itself to accommodate changing traffic demands, the time needed by service providers to turn up new services would fall from weeks, and even months, to mere seconds.

In addition to providing far more efficient network optimization, smart packet/optical interworking can greatly enhance network reliability. By enabling smart packet/optical interworking, service providers allow the layers to communicate. As a result, the packet layer can request additional optical bandwidth be allocated to a congested link, thereby alleviating the congestion and improving network reliability.

A second network reliability benefit enabled by smart packet/optical interworking is an integrated packet and optical response to abnormal network conditions. Today, when a network failure occurs, the packet and optical layers act independently to restore the failure.

The problem with this approach is twofold. First, the optical layer has no way to discern which traffic is worth protecting optically and which traffic is better left to a higher-level restoration scheme. Thus service providers that wish to employ optical protection must pay for optical capacity to protect premium traffic in addition to less important, best-effort services.

Second, in some cases, the packet and optical layer restoration schemes can actually work against each other. For example, in the case of an optical link failure both layers may attempt to create separate, alternative paths, thereby duplicating effort and wasting network resources. By allowing the packet and optical layers to communicate intelligently service providers can solve both these problems.

For instance, smart packet/optical interworking enables the optical layer to restore packet-layer traffic based on the traffic's class of service (CoS). Service providers can therefore save money by using APS to only protect premium traffic while leaving the restoration of best-effort traffic to packet-layer restoration schemes such as MPLS. In addition, the communication between the packet and optical layers ensures an efficient and coordinated approach to restoring for network faults.
Steve Garcia is senior product manager of IP product marketing for Brampton, Ontario-based Nortel Networks. He can be reached as sgarcia@nortelnetworks.com. Richard Kehler is senior manager of optical internet for Nortel and can be reached at rkehler@nortelnetworks.com. This article originally appeared on internetTelephony.com.

Visit the Nortel website.

RETURN TO TOP

---

From Amazon.com

Optical Networks : A Practical Perspective
by Rajiv Ramaswami and Kumar N. Sivarajan
$72.95; Availability: Usually ships within 24 hours.
Hardcover - 650 pages (February 1998)
Order this book.

Multiwavelength Optical Networks:
A Layered Approach
by Thomas E. Stern and Krishna Bala
$69.95; Availability: Usually ships within 24 hours.
800 pages 201 edition (May 7, 1999)
Order this book.

Understanding Optical Communications
by Harry J. R. Dutton
$64.99; Availability: Usually ships within 24 hours.
800 pages 1 edition (August 20, 1998)
Order this book.

RETURN TO TOP

---