Future optical transport networks
The article discusses the role of optical networks in the future and the new emerging techniques for construction of the network.
anders.berntson (at) acreo.se
andreas.aurelius (at) acreo.se
jonas martensson (at) acreo.se
The article takes a starting point in the requirements set by traffic and client networks. It describes the networking context and the relation to other networking technologies. Finally, the current trends and emerging technologies for constructing the optical network are discussed.
What will be transported through future networks?
The Internet traffic is growing quickly and Internet applications dominate the traffic volume in all network segments, access, metro and core. There is currently a strong demand for flat-rate Internet access and an increasing number of households are connected to the Internet through broadband access. The application that completely dominates the traffic is peer-to-peer file sharing (p2p). The figure above shows the result of measurements in a Swedish municipal network with more than 2000 end-users. Peer-to-peer file sharing applications, mainly Bit-Torrent, represent more than 90% of the traffic volume. Measurements in Acreo National Testbed show that p2p applications also have a high penetration, over 80 % of the households used p2p during the three months measurement period.
Are there any changes in sight? There are two applications or services that currently attract much attention, video and mobility. There seems to be consensus that the distribution of moving pictures will represent a dominating part of the traffic in IP-networks. The great debate is on the business model for delivering video: Should it be done
- with end-to-end service level agreement (SLA) similar to telephony
or
- without end-to-end SLA similar to the Internet?
A business model based on end-to-end SLAs requires additional systems e.g. for keeping track of application usage, for billing etc. Independently of the business model it seems clear that end-user applications will be IP based.
Mobile broadband where primarily Internet is delivered over mobile systems (3G) to computers or mobile phones is a big success. Functions for providing mobility are already implemented in the mobile systems. To the transport network the mobile networks represent a client network among others (perhaps with special requirements, e.g. to support synchronization of base-stations.) The bandwidth limitation imposed by the radio interface limits the forecasted traffic volumes of radio access networks to a fraction of that of fixed access networks. There is also within mobile systems, as within client networks in general, an ongoing migration to Ethernet interfaces.
Summing up, future networks will be dominated by Internet applications and forwarding of IP-packets. In addition, the networks should also deliver Ethernet services, transporting client Ethernet frames of different client networks. The Ethernet service traffic volume will be small compared with the IP-traffic volume. There is no demand for SDH services on the long-term, since also point-to-point wavelength services will mainly be used for transporting Ethernet frames. From the client perspective, future networks should provide cost effective packet transport.
How should a network be constructed to efficiently forward Internet traffic and deliver Ethernet services?
Since Internet will dominate the traffic, it is obvious that IP routers will be an essential part of future networks. It is also clear that optical transmission systems are needed to transport the traffic from one router to the next and they will be an equally important part of future networks.
It is still an open question how much functionality that should be put in different types of equipment. For instance, one way to deliver Ethernet services is by including additional functionality in the routers (MPLS together with PWE3 and L2VPNs) tunneling Ethernet frames over the IP network. For the optical transmission systems, new components such as the ROADMs and tunable lasers enable a reconfigurable lambda layer. The drive for introducing network functionality in the lambda layer comes primarily from the simplified operation brought by automated provisioning.
A basic network scenario that supports the requirements is then: IP routers capable of Ethernet service production via MPLS, PWE3 and L2 VPNs. The routers are connected by an optical network that is reconfigurable on the lambda layer.
If the network in the basic scenario would do the job, why should future networks not be built like this? The background to the question is that two new and competing layer 2 network technologies are today intensively discussed: PBB-TE and T-MPLS. Both technologies would be used for building an intermediate layer between the IP layer and the lambda layer, e.g. for traffic separation and traffic engineering. However, support of traffic separation and traffic engineering is not a unique feature of these two new technologies, it can also be done in e.g. SDH networks or IP networks. What is the edge of the new technologies compared to already existing alternatives? First of all, an SDH network is circuit switched and it was developed in a time when the traffic was dominated by telephony. The new network technologies are optimized for IP traffic and the term “packet transport” has been coined to emphasize this. Possible ways to provide more efficient “packet transport” compared with SDH, include statistical multiplexing, and the possibility to provide more accurate functionality, reducing complexity and cost of network management. “Packet transport” is about reinventing in a more cost-efficient way part of the SDH functionality.
But what about IP/MPLS as proposed in the basic network scenario? It is clearly a technology designed for IP-packets. What improvement would PBB-TE or T-MPLS bring when compared with IP/MPLS? This is a difficult question. Possible reasons could be cost savings obtained by by-passing IP routers for through traffic and more cost effective production of Ethernet services using native Ethernet rather then tunneling the Ethernet frames over an IP network. A more general formulation can be a more adequate selection of functionality. IP networks are highly functional and complex. Maybe it is a good idea to move network functionality to a simpler and less expensive network layer? (The next question, which I will not answer, is what functionality should be omitted?). However, the most important difference is probably not about functionality. All equipment vendors try to increase the value of their products. Router vendors can e.g. put colored interfaces directly on the IP-routers, which effectively integrate the optical network layer in their equipment. Optical network equipment vendors could increase the functionality in their equipment by combining the WDM-system (lambda layer) with layer 2 “packet transport”. This may reduce the need and the market for the IP routers.
Coming back to the original questions, first of how to construct a network that efficiently supports IP packets and Ethernet frames, the (very unoriginal) answer is: an IP-network of IP-routers, connected by a flexible optical lambda (WDM) network and with a logical and provisioned network layer in between the two. The choice of technology for the intermediate layer is under intense discussion presently. The difference between the alternatives is not mainly found in different functionality. The important choice to be made is in what type of equipment the layer 2 functionality should be included, in the IP-router (IP/MPLS), the optical network or perhaps separately (PBB-TE, T-MPLS)?
What is the future role of the optical network and which are the important new technologies?
Based on the above we argue that “packet transport” layer 2 functionality (PBB-TE or T-MPLS) is very interesting as add-on functionality in the optical network. The main objective, however, for the optical network is to provide cost-effective point-to-point transmission over long distances. The techniques for achieving this are currently developing very quickly driven by the success of Ethernet as a physical interface, by advances in digital signal processing, and by cost effective flexibility though tunable lasers and ROADMs.
When the standardization process for higher speed Ethernet started last year it became the starting point for development of matching optical transmission solutions. The next Ethernet bit-rate is expected to be 100 Gbit/s. Transmission at 100 Gbit/s compatible with the existing wavelength grid for WDM-systems raises a set of research issues, e.g. to reach high spectral efficiency and tolerance to fiber transmission impairments.
The second important factor that drives the development of optical transmission comes from the rapid development of digital electronics towards more complex electronic processing. The impact of digital signal processing is already vast e.g. by using forward-error correction (FEC) the required bit-error-rate has been shifted from 10-15 to 10-5 by adding 7% overhead. Digital signal processing can also be tailored to compensate for chromatic dispersion. The latest suggestion is to use digital processing to achieve phase recovery enabling coherent optical detection. With coherent detection both amplitude and phase of the received signal can be detected and used for carrying information. The spectral efficiency can be increased by sending multiple bits per symbol. It is also possible, in principle, to compensate electronically for all linear impairments of the fibre transmission. Dispersion tolerance becomes a question of computational capacity. However, a coherent receiver is polarization dependent. One way to address this problem cost-efficiently may be to transmit two signals in orthogonal polarizations for each wavelength and use polarization diversity in the receiver. Today there is no sign that this development should stop. Digital processing will enable 100 Gbit/s wavelengths with high spectral efficiency and with relaxed requirements on optical dispersion compensation using dual polarization, coherent detection and advanced modulation.
There is also presently a development from point-to-point systems to more advanced network topologies. The enabling components are tunable lasers and the wavelength-selective-switch (WSS). The WSS makes it possible to construct ROAMDs that are scalable from a single drop port to more complex multi degree nodes. The scope for a reconfigurable lambda layer is automated provisioning of wavelengths and simplified network operation. The tools that are needed for remote control of network elements can be provided by a management system or e.g. by the GMPLS controlplane.
The lambda layer is defined by the physical interfaces at the end-points of the lambdas. Reconfiguring the lambda layer means that additional end-points for lambdas will have to be introduced and new physical interfaces will have to be lit. Since transponders often dominate the cost in a WDM-system they are not installed in advance for future needs but when a new lambda is needed. This sets a practical timescale for provisioning of new lambdas to the order of days. ROADMs together with the GMPLS controlplane do not change this. However, ROADMs and GMPLS replace manual patching which e.g. increases the accuracy and reduces the errors in the process, it reduces the manual labour needed and it enables keeping accurate record of the network state.
Traffic engineering on the lambda layer is still possible but it occurs on a long timescale in response to long-term traffic variations. A provisioned network layer can be reconfigured on a short time-scale without changing the hardware, provided that
- it supports sharing of physical interfaces and,
- there is spare bandwidth within the existing hardware.
This is normally not the case for the lambda layer but it is true for the layer 2 packet transport. It sets the scope for the layer 2 packet transport to make efficient use of the underlying lambda network and to provide traffic engineering in response to relatively short changes in the traffic load.
Conclusions
We have argued that the future metro/core networks will consist of IP-routers connected with a reconfigurable WDM-network. The scope of reconfigurability of the WDM network is automated provisioning of new lambdas to meet long term changes in the traffic load. The WDM layer will carry 100 Gbit/s Ethernet signals with an order of magnitude higher spectral efficiency than today thanks to advances in digital signal processing and modulation formats. Between the IP-layer and the WDM-layer there should be an intermediate network layer for packet transport for the sake of logical traffic separation and for traffic engineering in response to changes in the demand on a relatively short time-scale.
For this there are three main alternative network technologies: IP/MPLS, PBB-TE and T-MPLS. The choice among the three is a matter of selecting in what type of equipment to put the layer 2 packet transport, in e.g. an IP-router or in the optical network.
Definitions
WDM (Wavelength Division Multiplexing) is a technology that enables combining several optical data signals having different wavelengths in the same optical fibre. WDM-systems are constructed for data transmission over long distances. ITU has standardized some aspects of the WDM, e.g. the wavelength grid.
ROADM (Reconfigurable Optical Add Drop Multiplexer) is a network element of the WDM system where it is possible to dynamically and remotely select which lambdas should be added/dropped and which should be passed through.
MPLS (MultiProtocol Label Switching) is an Internet Standard where forwarding decisions of IP routers are based on an additional packet header, the MPLS header containing an MPLS label. Compared with an IP address which identifies the network destination, the MPLS label is only of link local significance and can be swapped at each hop. Forwarding tables in MPLS routers concatenate links to form a path though the network. MPLS enables traffic engineering and traffic separation in IP networks.
PWE3 (Pseudo-Wire Emulation Edge-to-Edge) is an Internet Standard for encapsulation and point-to-point transport of data other than IP packets, e.g. Ethernet frames, through so called pseudo-wires over a packet switched network (e.g. an MPLS network).
L2VPN (Layer 2 Virtual Private Networks) is a set of IETF standards for proving end-to-end layer 2 services through an IP network using pseudo-wires. Forwarding decisions taken by the service provider are based on layer 2 client information. The Virtual Private LAN Service is one version of the L2 VPNs that emulate a LAN service e.g. so that client devices belonging to the same VPLS appear to be on the same bridged Ethernet.
PBB-TE (Provider Backbone Bridging – Traffic Engineering) is basically a new version of Ethernet that is standardized by the IEEE. Compared with traditional Ethernet the functions for flooding, learning and the spanning tree protocol have been switched off. Forwarding tables are instead configured using management system or a control plane technology.
T-MPLS (Transport-MPLS) is standardized by the ITU. The idea behind T-MPLS is to make MPLS a separate set of functionality that can be used independently of the IP-routers and for other types of equipment.
GMPLS (Generalised MPLS) is a set of IP protocols standardized by the IETF defining a control plane by generalizing the MPLS paradigm to non-packet label switched paths, e.g. time-slots, wavelengths etc. Of particular interest at the moment is the application and adoption of GMPLS for lambda switched networks (ROADMs) and for Ethernet (using GMPLS for PBB-TE).