The ALL-OPTICAL IP network

DTM could be the answer to delivering true QOS in fiber optic networks

New network architectures can let carriers reduce their up-front expenditures and decrease the complexity of the overall network design, which helps them keep a firm grip on future maintenance costs. But cutting costs is not the only reason for rethinking network architectures. Fast provisioning of new revenue-generating services that limits the risk of network congestion is equally important.

Two significant trends affect the way green-field service providers design their networks: IP and fiber optics.

First, IP is becoming the prevailing protocol for a variety of services. As the Internet grows, new services are developed at a pace that the industry has never seen before. In addition, real-time applications such as voice, streaming video and audio have been introduced for IP. The Internet eventually will evolve to a full-service network. As a result, few, if any, services are developed for native ATM, frame relay or Sonet architectures.

Second, new techniques for deploying fiber optic infrastructure cost-effectively has changed the available capacity in the networks. The advent of advanced fiber optic technologies such as dense wave division multiplexing (DWDM) increases the available bandwidth beyond terabit rates in existing fiber optical links by using different wavelengths. Now, the capacity growth in fiber optic transmission links is a hundredfold faster than the capacity growth in switches and routers, which should substantially affect the way future high-speed networks are designed.

DTM to the rescue

Ever since the Internet revolution changed the communications industry in the early 1990s, ATM has been trying to catch up to the requirements in the IP centric world; for example, by introducing multiprotocol label switching (MPLS) and reducing cell tax.

Still, many carriers and service providers want to decrease network complexity by reducing the number of managed layers. It is less complicated to use IP routers to build an IP network than to use ATM switches and routing protocols at the ATM and the IP layers. To address this, packet over Sonet was introduced as a simplification. However, the static nature of Sonet and the high bandwidth offering when using DWDM puts high stress on routers, which in turn provides opportunities for new technologies that more efficiently map the IP traffic on wavelengths. Dynamic synchronous transfer mode (DTM) is designed to meet the requirements of the optical internetworking era (Figure 1).

DTM is a dynamic time and space division multiplexing architecture that provides a thin link layer for mapping IP to the optical layer. The protocol is optimized to carry IP traffic and includes only what is missing in IP - such as quality of service (QOS), fast redundancy and efficient multicasting. No feature that already is present in the IP layer is doubled.

The overall objective has been to provide a simple solution for all-optical IP networks. Features include:

- Support for strict QOS

- Distributed network architecture that offloads routers in an optical ring deployment

- Scaling to hundreds of gigabits per second using standard hardware

- Flexible billing models enabled by control of network usage upstream and downstream.

Never too strict

Two major demands push the industry to introduce strict QOS in IP networks.

First, to accommodate real-time services, there is a need for a technology that can reserve resources between the sender and the receiver. Real-time applications need a smooth stream of data end to end. Voice traffic traditionally uses circuits to ensure this. When moving to the packet-oriented data world, service providers need to guarantee the quality of the network service transporting the flow of data over the network.

Second, service providers should be able to sign particular service level agreements (SLAs), guaranteeing certain service levels independent of the traffic situation on the public infrastructure. To guarantee a service level implies control of network resources and the ability to log usage data, link capacity and overall availability. As mission-critical applications move from the LAN to the WAN environment, quality agreements become even more crucial.

Why should a company using frame relay for LAN interconnect change to an IP-based virtual private network solution if there is a degradation in quality? Why would anyone replace existing PBX and telephone equipment if the quality of voice over IP can't be assured?

Several technological advances address these questions. DiffServ or TOS, resource reservation protocol (RSVP), MPLS tagging and various forms of traffic shaping and policy functions such as common open policy service are some of them. The list increases monthly.

The problem is that prioritization and policy functions don't provide any strict guarantees. Instead, they ensure that prioritized traffic will be first in line if congestion occurs. It is complex to keep track of the aggregated level of prioritized traffic in a large network. What happens when large amounts of prioritized traffic from different sources end up in the same router? Congestion, added latency and eventually buffer overflow will occur. Moreover, advanced prioritization solutions require complex buffer handling and decrease the overall network efficiency.

Instead of trying to achieve the QOS capabilities from the complex top-down approach, it would be tempting to design from the ground up. During the evolution of packet-switched networks, the inherent and indisputable advantages of circuit-switching became thought of as inefficient and cumbersome. However, does a more efficient technology exist for true QOS than time division multiplexing (TDM) circuits?

DTM in the network arena

So, what if service providers could offer true QOS without the disadvantages of traditional circuit-switching? What if providers could remove the constraints of static circuits and heavy nationwide signaling but still have no-compromise QOS? What if vendors could make a distributed network architecture that inherently supports QOS for IP traffic? These were the questions when DTM was developed and entered the network arena.

DTM was developed by applying an IP-centric approach to Sonet. It takes advantage of the best in Sonet and synchronous TDM technologies, such as inherent traffic isolation, low latency and predictable traffic. However, Sonet was developed to transport voice traffic with symmetrical traffic patterns - 64 kb/s in each direction.

This is not the case for most IP applications. Instead, the traffic pattern can differ from real-time critical streams of IP packets to bursty file transfers. DTM uses flexible circuits. These circuits are referred to as channels that can perfectly match the customer's needs or even the need for a single application, depending on requirements. Channels are set up on-the-fly in less than a millisecond.

DTM provides channels across the network and enables traffic shaping at high connect rates. The channels are set up in linear steps of 512 kb/s to offer jitter-free connections across the network, which allows for toll-quality voice and delivery of high-quality video services. The delay is kept constant and low because there are no packet buffers within the DTM network. Thus, the DTM network inherently supports QOS at the optical transport layer.

How does the network equipment get the information for which bandwidth settings should be used for a specific customer or application? Here, all initiatives at the IP layer are used to give the hard resource allocations at Layer 2. MPLS tags can classify traffic and are used to decide the characteristics of the DTM channel to set up throughout the network. In fact, MPLS tags map the channel IDs. A DTM-empowered IP router can use information from Layer 2, 3 or 4 to classify traffic and allocate channel resources corresponding to the traffic needs.

How it works

Filters map IP packets onto DTM channels with certain characteristics. The filters have an open application programming interface toward different classification and policy sources. Based on the sender or receiver address, port number, DiffServ field or RSVP information, channels are set up with different bandwidth and priority characteristics. Channel specifications are grouped into accounts that typically map the SLAs signed by the customers. Certain accounts can be given priority over best-effort traffic on the network.

For example, if there are free resources, best-effort channels are dynamically added along with the changes in traffic patterns, and statistical multiplexing is achieved onto these channels to maximize bandwidth use. When a channel with higher priority is being set up, and there are no free resources, the best-effort channels are preempted to free resources for the prioritized channel (Figure 2).

DTM-empowered IP routers integrate the functionality of an add/drop multiplexer (ADM), ATM switch and an IP router, reducing the number of network devices and thereby limiting investment costs. The linear multiplexing scheme and dynamic features in DTM allow for different service offerings and QOS beyond today's costly ATM solutions (Figure 3).

DTM is optimized for IP and provides plug-and-play operations via capabilities such as globally unique, permanently assigned IEEE MAC addresses and automatic topology discovery through the auto-configuration protocol. The routers also enable comprehensive, integrated, SNMP-based management capabilities.

The all-IP approach also enables the service provider to simplify maintenance and support. Because of the decreased number of network layers, management need only take place at the IP layer. Standard techniques are used for distributing routing and network policy information. And no extra global addressing is needed nor complex routing protocols running in parallel with the well-known and accepted routing protocols for IP traffic.

DTM uses a distributed routing scheme, meaning that not all channels are terminated at every node when deployed on a shared medium such as a ring or a bus. Instead, only the channels that contain traffic destined to the node are terminated there. The node will process only traffic that is being routed or switched by the node. This means that the routers connected to the network will be offloaded compared to, for example, packet-over-Sonet networks in which the full capacity of the link is terminated in the router. With development in fiber optics and DWDM, the routers evolve as the bottlenecks in next generation networks. The distributed architecture becomes an important feature to scale to OC-192 and higher speeds.

The distributed architecture affects future product enhancements and, more interestingly, it most significantly simplifies network planning. DTM networks can be planned on an aggregated level because only the amount of traffic passing through a node onto the shared medium needs to be calculated.

The bandwidth between the routers automatically is changed according to the actual traffic pattern. New revenue-generating services, therefore, can be provisioned with reduced risks of network congestion caused by a sudden increase in bandwidth demand. Furthermore, in a distributed architecture, there are no costly centralized routers that need to be upgraded when the network grows geographically with new users.

An elegant solution

The inherent simplicity and predictability of the DTM technology allows for scaling to OC-768 and beyond using simple and standard hardware. Today's implementations support 10 Gb/s without the use of any application specific integrated circuits. Keeping things simple not only helps when it comes to network scalability and planning, but it makes the solution more reliable, limiting costly downtime.

Emerging carriers and service providers can carry out an all-optical IP network and thereby leapfrog with advances in fiber optics and IP routing architectures. This can be achieved by investing in equipment optimized for IP, rather than using unnecessary ATM switches, Sonet ADMs and digital cross-connects.

DTM is a distributed broadband network architecture that provides a thin and efficient link layer for optical IP networks. It scales to unsurpassed speeds and offloads the routers when deployed in shared medium implementations, ensuring flexible network planning and reduced risk for network congestion. Using DTM-empowered routers brings QOS to IP networks without adding the complexity of ATM or the static behavior of Sonet.