All optical networking?
Many industry commentators seem to believe that end-to-end optical networking is the future. But paring down the network to a single layer compromises its intelligence. Next-generation networks will instead depend on a two-layer architecture - using both IP and optical protocols.
By Alan Taylor, Technical Director EMEA, Juniper Networks
While traditional network architectures have comprised four layers - the physical fibre, the optical multiplexing layer, a switching layer (ATM) and an IP routing layer - many service providers and larger enterprises are now responding to changing market conditions in order to optimise operation and maintenance costs. The rise of IP as the dominant networking protocol, advances in transmission technologies and the pervasive influence of the Internet, mean that service providers are consolidating their infrastructures wherever possible.
The industry is reducing this layer count by transferring the benefits delivered by ATM on to the IP layer through developments such as MPLS. In the largest IP networks the ATM layer is already disappearing because IP routers now have improved packet processing and throughput speeds. The preconception that IP is merely a "best effort" non-deterministic transport mechanism is no longer accurate. IP is deterministic, reliable and robust enough for carrier-grade networking and does not suffer from the scalability constraints suffered in ATM networks.
No single-layer networks
Some people argue that the natural outcome of network consolidation is a single, all-optical layer, but there is little of substance to support this discussion. Although optical vendors have made tremendous strides in the way paths are created, optical networking focuses on manipulating these paths in a fast, flexible way. None of the industry's activities concern the content and behaviour of packets - this is left to the IP routing layer.
An all-optical network envisages the ability to make decisions about each individual packet while the packet is still within the light-stream. In reality, the optical domain has no idea even what a packet is or what it means: commercial lambda, wavelength and channel switching devices currently revert to the conventional electrical domain in order to switch the traffic to a different wavelength.
The necessary network intelligence must come from IP routers, which not only switch packets between networks, but also contain packet-processing tools needed to create IP services. Packet processing involves removing the data link encapsulation from each packet then inspecting and perhaps modifying the network layer packet header fields.
Routers can identify packets that need to receive specialised handling. The fundamental packet processing capability of routers allows service providers to create and deploy revenue-generating IP services. Optical cross-connects (OXCs) cannot recognise individual packet boundaries, so they cannot perform the packet processing needed to support such services.
To support the idea of packets without converting the stream from optical to electrical domains demands optical logic: gates, equivalent to those in electronic transistors. Their development will take many years to match the miniaturisation and mass-production techniques of conventional technology.
Routers also support statistical multiplexing of traffic flows - essential to the efficient operation of extremely large IP networks. Because statistical multiplexing creates the likelihood that a particular link might occasionally be oversubscribed for a short period of time, packet buffering must be available for each router interface. Since data applications are bursty by nature, a relatively small amount of packet buffering can deliver a significant gain in network operational efficiency. Technological advances in the foreseeable future will continue to support electronic but not photonic memory subsystems, so OXCs will not be able to provide the buffering capabilities needed for the statistical multiplexing that are essential for efficient network operation.
Likewise, optical memory is in its formative stages - light can be slowed down a little, but not yet stored successfully.
The two-layer approach
The optimum network is based on a two-layer architecture (see diagram below) - a routing IP layer and a transmission optical layer. Here, routers make decisions about packets, understanding where they have to go, whether to block them and so on; the transmission layer provides flexible connection paths between the routers.
Transmission nodes such as OXCs handle the switching of whole fibres, individual wavelengths or even time-slots within wavelengths, if SDH functionality is integrated within them. Above these devices are the routers connected by whatever topology the carrier chooses. This model is both logical and inevitable because both functions - optical and routing - are fundamental and belong to separate domains.
Linking the two layers
To link the IP and optical layers, the IETF (Internet Engineering Task Force) has defined a model based around Generalised Multi Protocol Label Switching (see box, right).
GMPLS enables service providers to make use of any traffic format in the forwarding plane including wavelengths, time-slots or fibres. Here, the IP/MPLS layer functions as a peer or equal of the optical transmission layer: a single routing protocol (IP) runs across both domains.
Each GMPLS-enabled node exchanges routing information with a neighbour and it is this feature that allows the routing protocols to scale to support the creation of extremely large IP networks, while making decisions based on user preferences.
The use of a fully integrated routing protocol also simplifies the provisioning of wavelengths across the optical domain. It supports rapid automated recovery in the event of equipment failures and makes more efficient use of underlying optical network resources.
According to the Yankee Group, Internet traffic is currently growing at 80% per year. The relationship between IP and the underlying optical transmission network is changing. The early stance - that IP is simply one of many service types that run across the transmission network - must be updated. A combination of optical and IP infrastructure, delivering paths and intelligence respectively, will underpin the delivery of revenue-generating services and will form the basis of the new public network.
Benefits of Generalised Multi Protocol Label Switching
GMPLS provides a powerful and flexible solution for signalling and routing in the new IP infrastructure. Support for open standards allows carriers and service providers to select best-of-breed equipment as they continue to build out their networks. The peer model allows IP routers to make more efficient use of optical network resources when calculating the various physical paths.
By removing the overlay model's requirement for an "n-squared" mesh of optical channels to support the exchange of routing information, the peer approach enhances the scalability of IP routing. GMPLS leverages existing provider and vendor operational experience with MPLS traffic engineering.
Furthermore, GMPLS eliminates the need to reinvent, test and qualify an entirely new class of control protocols. Open standards promote the parallel evolution of standards for existing multi-protocol and for future all-IP networks, so they can continue to support rapidly changing provider and subscriber requirements. GMPLS gives providers the flexibility to deploy the proper tool for specific application needs. GMPLS makes possible the rapid development and deployment of a new class of OXCs that facilitate provisioning on demand.
Europe's R&D Géant demonstrates Juniper's IP-optical approach in practice
Dante (www.dante.net) has developed a pan-European research and education network, Géant, that connects the national research and education networks of more than 30 countries (see July issue, page 33). Funded by the European Commission, the network enables over 3,000 research and education institutions to share vital information concerning research activities, including new grid development projects such as DataGrid and EuroGRID which manage large distributed data sets and distributed computing for weather forecasting, respectively.
Experience with Géant's predecessors, such as the TEN-155 network, showed that, as demand for the network increased, bottlenecks began occurring and quality of service could not be guaranteed. Higher transmission speeds were required, and this was one of the objectives of the Géant project.
The network runs IP over a mainly optical infrastructure to create a two-tier network which offers significantly increased bandwidth compared to its predecessors, resolving capacity problems along the pan-European backbone of the network.
Dante chose Juniper Networks to implement IP backbone routers as part of the Géant architecture. The resulting infrastructure offers greatly increased capacity, resilience and speed across all 30 countries, providing the European research community with an efficient infrastructure. Juniper's IP routers are deployed in the main points of presence in the Géant network - Vienna, Brussels, Geneva, Prague, Frankfurt, Madrid, Paris, Budapest, Milan, Amsterdam, Poznan, Stockholm, Brastislava and London. Juniper's systems allow Dante to maximise the routing capabilities and ensure QoS and bandwidth. The infrastructure operates nine trunk circuits at 10Gbit/s and a further 17 operate at 2.5Gbit/s. Connecting the POPs in cities to several international circuits in different countries has raised network resilience.
Dante has leased two independently routed, unprotected wavelengths into each PoP, to ensure resilience and high availability. This, combined with the ability of the routers to re-route IP traffic in the event of a link failure, ensures that the research bodies receive a high-quality service. Having just two network layers (IP and optical) simplifies the management of the network. Since the network was first installed in December 2001, it has operated reliably and predictably, claims Juniper.
Technical Director, EMEA