- This event has passed.
“Mathematical Techniques in Optical Networks”: Introduction and Scope
Apr 16, 2018 - Apr 18, 2018
Optical networks are a key part of current day technological progress. They form the backbone of all future communication networks like 5G, internet of things and such. Optical Networks form a very important field of research in engineering, and also pose a lot of interesting mathematical problems.
The workshop Mathematical Techniques in Optical Networks is conducted under the auspices of NETWORKS and The Research Center for Integrated Nanophotonics at Eurandom. The goal of the workshop is to bring together the communities of engineering and mathematics. The event will be set up in such a way that researchers, especially young mathematicians, receive tutorials and lectures from experts in the field of Optical Networks. The speakers will address challenges in the existing technology and ways to broaden the scope of these networks through mathematical tools.
The workshop consists of talks and tutorials by leading researchers in the field who will cover topics like architecture, stochastics, optimization, machine learning, in the context of Optical Networks. There will also be several timeslots for contributed talks by young researchers working on similar problems, and a poster session.
Abstracts for presentations:
Participants interested in presenting their work, can send their abstract to the following e-mail address:
Nail Akar, Bilkent University
Annie Gravey, Telecom Bretagne
Frank Phillipson, TNO
Mario Pickavet, Ghent University – IMEC
Marco Ruffini, University of Dublin
Massimo Tornatore, Politecnico di Milano
Cedric Ware, Telecom ParisTech
Titles + Abstracts
Murtuza Ali Abidini
Resource allocation in an optical router
In a telecommunication network, packets have to be routed from source to destination, passing through a sequence of links and nodes. Packets from different sources are time-multiplexed and thus flow sequentially through the network’s links. When arriving at a routing node, they need to be queued in a buffer, where they need to wait before they can be forwarded to the appropriate outgoing port of the node and travel further through the network. This store-and-forward procedure can cause a serious increase of the latency, and increasingly so when the traffic load in the network grows. Hence the buffering processes need to be designed as efficiently as possible. The performance analysis of optically-routed networks brings additional challenges with respect to the analysis of networks which deploy electronic routing.
In this talk, we will discuss a revenue (or throughput) maximization problem for optical routing nodes. We model the routing node as a single server polling model with the aim to assign visit periods (service windows) to the different stations (ports) such that the mean profit per cycle is maximized.
Under reasonable assumptions regarding retrial and dropping probabilities of packets, the optimization problem becomes a separable concave resource allocation problem, which can be solved using existing algorithms. Then we extend this to a multiple wavelengths node and present an efficient and accurate heuristic procedure for solving the NP-hard revenue maximization problem and investigate the advantage offered by having multiple wavelengths.
Performance Models for Optical Burst/Packet Switching Systems
To cope with bursty traffic and for more efficient utilization of the fiber capacity, two packet-based optical switching paradigms, namely Optical Packet Switching (OPS) and Optical Burst Switching (OBS), have been proposed. This talk addresses the tele-traffic modelling of such systems in asynchronous mode of operation with variable length optical packets (or bursts). One of the major issues in optical burst/packet switching nodes is contention which arises as a result of two or more incoming optical packets contending for the same output wavelength. Since buffering is not available in optical networks in the usual sense, such contention is resolved either in wavelength domain by wavelength converters (WC) or in time domain by fiber delay lines (FDL). This talk will overview the performance modeling of contention resolution mechanisms and also QoS differentiation mechanisms in packet-based optical networks. As the mathematical instrument, the emphasis will be on multi-regime Markov fluid queue models and their applications to optical networking.
Throughput Gains in Optical Networks via Adaptive Transceivers
In this talk, the potential throughput gains that can be achieved by using adaptive transceivers are discussed. We use an information theoretic-approach and show results for different network topologies, including the Deutsche Telekom Germany network, the NSF Mesh Topology, and the Google B4 Data Center Network. The analysis also considers the tradeoffs between hard-decision and soft-decision forward error correction strategies.
A retrial queueing model for optical packet switching
In optical packet/burst switching, fibre-loop optical buffers provide a compact and effective means of contention resolution. In case the optical packet length is fixed, it is a natural assumption to choose the optical loop length equal to the packet length. When an infinite number of loops are arranged in parallel, such an optical buffer can be modeled by means of a retrial queueing system with deterministic retrial times. While the availability of gaps between the transmissions in an optical buffer typically reduce the available throughput, we show that for these fibre-loop buffers, this is not the case if one allows that the packets are reordered. Somewhat surprisingly, this observation follows from a correspondence between time-discretised retrial queueing systems and exhaustive polling systems. This correspondence can further be exploited to study performance of the (optical) retrial queueing system.
Modelling Contention Avoidance and Resolution in Optical Networks
Optical networks present at the same time a very large capacity and very poor buffering capabilities. This is why they are currently deployed as bufferless systems both in the transport network, where fixed capacity circuits are statically set up, and in FTTx access networks where traffic is gated in the electronic stratum and optical bursts are scheduled by a single controller. However, a rich corpus of research has been devoted in the last 20 years on how to harness the huge optical capacity to efficiently support high bit rate bursty traffic.
The first approach relies on optical buffers provided by Fiber Delay Lines (FDL) in order to delay information within the optical layer in case of contention. Delaying information thanks to a FDL drastically differs from buffering a packet performed within the electronic stratum. Indeed, when information is sent on a FDL, it is simultaneously scheduled for transmission at a time specified by the length of the FDL, whereas a packet stored in an electronic buffer can be scheduled by an independent process, that can take account of events occurring after it has been stored. Also, a tagged optical packet contending for transmission on a link can be blocked either by another packet which is currently being transmitted, or by a packet currently delayed in a FDL, which is scheduled to be transmitted during the transmission time of the tagged packet.
The second approach assumes a bufferless optical network and relies on grooming the traffic within the electronic stratum, and on controlling its transmission as an optical packet in order to limit, or to completely avoid, losses within this layer.
With both approaches, several system characteristics have to be taken into account. An important one is the number of optical channels made available to the system. Multiple channels allow load balancing and thus naturally limit contention. For example, the deflection of an optical burst from one wavelength to another is the typical implementation of a “hot potato forwarding” strategy in optical networks. Another characteristic is whether the system is synchronous or asynchronous. Synchronous operation limits contention but may not be simple to enforce globally, except on a specific topology, e.g. a tree (as in FTTx) or a ring. Lastly, how the optical burst is built also impacts on the performance delivered to electronic traffic.
In this presentation, various modelling techniques used to assess the performance delivered by optical systems or networks, shall be presented. As pointed out previously, dealing with contention in the optical layer is quite different from dealing with contention in the electronic stratum, and approaches fitting the latter do not directly apply to the former.
Dynamic optical networking in the user access domain
Optical networks offer extra dimensions for network reconfiguration and traffic routing, such as the wavelength domain and the spatial modes domain. These extra degrees of freedom are not only very valuable at the higher hierarchical levels of networks, but also at the ones very close to the users. Some thoughts will be shared about the benefits in fibre-to-the-home networks, fibre-in-the-home networks, and optical wireless indoor networks.
Next-Generation Optical Access Networks: the Energy Challenge
The success of novel mobile broadband Internet services is leading to an enormous increase in the traffic transported by optical networks at different levels, ranging from core networks, down to metro and access-aggregation sections. In particular, next generation (5G) mobile access will revolutionize current optical access-aggregation networks, targeting unprecedented performance not only in terms of higher data rates per user and lower latency, but also in terms of network intelligence, and capillarity. Therefore, the supporting optical access-aggregation network, which represent a major contributor of global energy consumption, is expected to adapt to this drastic evolution. In this talk, we focus on the Centralized Radio Access Network (C-RAN) architecture, and identify the energy benefits obtained through centralization of baseband processing into shared locations. We show that the arising multiplexing gain results in substantial reduction of the required housing facilities, network devices and signal processing resources, thus enabling significant cost and power savings.
Performance and modeling of Metro Networks for 5G applications
The next generation mobile communication system (5G) will operate in a highly heterogeneous environment characterized by the existence of multiple types of access technologies, multiple types of devices, multiple types of user interactions, etc. To support traffic originated from heterogeneous 5G access networks with its specific use case, a flexible optical metro network infrastructure is necessary. The metro network node should combine the novel SDN, NFV technologies and elastic cloud services to implement slice-able network for the specific use cases in 5G network. In this talk, I will give an introduction of the metro node architecture and the simulation results of its performance in network slices layer. For the simulation, several 5G applications’ traffic pattern have been modeled, and the performance of latency and packet loss in each slice have been analyzed.
Smart City Services Infrastructure Planning
In Smart Cities many services or monitoring systems will arise, like security cameras, air quality and pollution monitoring systems, event detection systems, internet access and communication systems for smart vehicles enabling congestion control, smart parking and (smart) road condition systems. These applications assist in the transition to ‘smart cities’ for which the ultimate goal is to improve the quality of life in a city. However, where mobile telecommunication operators are already limited in the search for locations for their equipment, in the near future this will be worse. A potential solution will be the integration of these locations with (already existing) street furniture like lamp posts and bus shelters, which are a dense infrastructure and are close to the place people live and work. Then the question arises how to choose the right locations (of the many options) to deliver the required coverage of all the various services at minimal costs.
Modeling of Passive Optical Networks
Passive Optical Networks (PONs) are generally considered as an important technology to build high capacity access networks and to bring the fiber closer to the user. In his talk, Mario Pickavet will elaborate on some challenges and potential solution methods when modelling these PONs. Design of an efficient network layout will be tackled. Also medium access control protocols for upstream traffic will be investigated and some queueing models will be discussed.
Load-aware resource allocation in dense Radio-over-Fiber wireless networks
The wireless networks providing us with WiFi, 4G and in the future 5G connections have reached the limit of their capacity due to the massive number of wireless devices and bandwidth-hungry applications. To increase the capacity of these networks, operators install more base stations in an area, allowing for smaller coverage areas and frequencies to be reused more often. This network densification introduces new challenges in the area of resource allocation.
In this talk I will describe a system that combines wireless and optical communication, so-called Radio-over-Fiber networks, and explain how these two domains influence each other and give rise to new frequency allocation constraints. Lastly, I will show a dynamic frequency allocation algorithm which uses load measurements at the base stations. I will present numerical results that show how different parameters influence the behaviour of the algorithm, and also that this dynamic algorithm outperforms a static allocation.
The virtualisation of telecommunications networks and its effect on optical networking
Starting from a small university-led project, OpenFlow has in less than a decade revolutionised the way networks operate. The simple idea of separating the control plane from the data plane has quickly evolved into more complex concepts of network function virtualisation and multi-domain multi-layer network control and orchestration. Following on from this concepts, telecommunications operator have recently considered the possibility of virtualising their central offices, taking part in standardisation and open source frameworks trials for network virtualisation, such as CORD and OPNFV.
The talk will discuss how this brings both unprecedented opportunities and challenges to the optical transport network, especially in the access/metro region. On the one hand the use of truly dynamic wavelength allocation and transceiver reconfiguration, can reduce end-to-end latency and assure higher levels of quality of services, thus satisfying the requirements of next generation of services and applications. On the other hand the types and amount of resources to be orchestrated increases exponentially, making its implementation difficult to scale.
Towards Flexible Passive Optical Networks
Current passive optical networks (PONs) are fairly static and do not cope with traffic changes. Hence, dynamic bandwidth allocation is necessary to meet users demands. This talk will use as an example a deployed PON and will discuss how to introduce flexibility in such a PON.
Emerging machine learning applications in optical transport networks
Despite indisputable progress towards higher capacity, management of optical networks (ONs) still requires complex and time-consuming human intervention. Machine Learning is currently being investigated as a promising solution to automate ON management. In this talk, I will introduce some applications of machine learning in ONs, with a focus on QoT estimation and failure identification.
Virtual Channel Transmission for Zero-Jitter Deterministic Scheduling in WDM Slot Switching Xhaul
In this presentation we discuss the support for deterministic scheduling in WDM Slot Switching Xhaul. In WDM Slot Switching network called N-GREEN the WDM slots carry the payload data units (PDUs) over parallel wavelengths, in a time slotted manner. Such network design enables savings in cost and energy consumption w.r.t. the state of the art electronic packet switching technologies through the optical parallelization. The subsequent time slots in N-GREEN are separated by a fixed guard time interval, to achieve the separation and switching of individual WDM slots. We show that under this assumption it is possible to provide zero-jitter transmission of isochronous traffic, that is present in 5G Xhaul, irrespective of the size of the guard time. In finding the solution, we first observe the properties of isochronous traffic, and we then introduce “virtual channel transmission” for WDM Slot Switching network as a new method for enabling zero-jitter transmission of time-sensitive traffic in this network segment. The scheduler is first implemented by using linear programming and next, a scalable implementation of the same scheduler is presented in form of a greedy algorithm. The numerical results confirm the cost and energy consumption savings of N-GREEN network, when compared with Ethernet state of the art solution, for time-sensitive traffic, while zero-jitter deterministic scheduling is achieved. We discuss also the perspectives of this work and show preliminary ns3 simulation results for a more detailed performance evaluation and advanced schedulers in N-GREEN.
(joint work with Annie Gravey, Dominique Chiaroni, Brice Leclerc, Thierry Zami, Philippe Gravey, Michel Morvan, Dominique Barth and Djamel Amar)
Kurt Van Hautegem
Selectively delaying packets in scheduling algorithms for OPS/OBS networks
With ever-increasing demand for bandwidth, both optical packet switching and optical burst switching are proposed as alternatives to increase the capacity of optical networks in the future. In these packet-based switching techniques, either Fiber Delay Lines or fiber loops are used to avoid contention between packets on a single wavelength. The involved scheduling algorithms decide on which Fiber Delay Line or fiber loop each packet is scheduled in order to maximize performance. By selectively delaying packets longer than strictly necessary, our proposed scheduling algorithms can outperform existing scheduling algorithms by tens of percentages in a variety of settings. In this talk we will focus on how the concept of delaying packets longer than necessary can be implemented in a specific setting and go into more detail on the performance results obtained by Monte Carlo simulation.
Future trends in optical networks—and why you should care
The ongoing digital revolution continues to drive and rely on a massive growth of ubiquitously-available data connectivity. Unfortunately, current networks may not be able to meet the required capacity, as they already consume a significant and exponentially-increasing portion of the global electricity supply; and new latency constraints don’t mesh with current network paradigms. Optics, as the current medium for most data transmission, can bring substantial contributions to these issues. However, using it as anything other than the current “fat pipes” may require going beyond the layered-network paradigm. After a brief overview of what makes optics so special, we will focus on its current use and likely short-term evolutions, then point out a few research directions as concrete illustrations of cross-layer network optimizations.
Modeling and performance investigation of high performance data center architectures based on fast optical switches and fast flow control
Cloud applications, big data, internet of thing are massively increasing the traffic within data centers and putting stringent requirements on the data center network (DCN) in terms of connectivity, bandwidth, latency, costs, and the energy consumption. There is a call for the architectural and technological changes of data center (DC) to meet the high performance of the traffic demands and the capital and operation cost effective.
In this talk, we will talk about a novel DCN architecture HiFOST based on fast optical switch (FOS). The functional blocks of HiFOST and the operation details of HiFOST will be discussed. To investigate the performance of HiFOST, we have modeled the network components including server, top of rack switch (TOR), transceiver, FOS, et.al in the OMNeT++ simulator. There are several major characters in our network model. First, a realistic DC traffic model with handover between ON and OFF period was implemented. The Pareto distribution is used to model the length of the ON/OFF period. While the packet length varies from 64 to 1518 bytes which is in accordance with the real traffic. Second, packet segmentation and assembling mechanism have been implemented at the modified TOR. Packet are segmented into cells with 64-bytes when buffered at the TOR and 5 cells are assembled into 1 optical packet before sending out. Third, the modeling of the broadcast and select FOS was discussed in details. Fourth, a wavelength mapping equation is proposed to avoid the wavelength overlap at the FOS. Besides, the fast flow control for the FOS was carried out to solve the packet contention happened at the FOS. The operation and the modeling of the flow control mechanism will be discussed. In the end, the performance results in terms of packet loss, latency and throughput will be given under different buffer size, link capacity, and traffic pattern.
Modeling QoS in Multi-class OBS/OPS Systems
Horizon-based reservation in OBS systems is an early reservation scheme in which an optical node keeps track of the remaining time until it will become free to forward an incoming burst, which is named the “horizon” parameter. A burst control packet (BCP) is sent to the optical node before the arrival of each burst to indicate the offset time the corresponding burst will arrive. In this way, an optical node can decide whether to accept a burst or not by looking at its offset time. If the offset time is less than the horizon, the burst is rejected. Otherwise, it is accepted and the horizon parameter is updated accordingly. In this study, we model the horizon-based reservation in a two-class OBS system using Markov fluid queues. In light of this model, we investigate offset-time based QoS differentiation in such systems and show that deterministic offset times (as opposed to exponentially distributed) lead to lower burst blocking probabilities as well as better QoS class separation. In the second part of the study, we model a two-class OPS system in the presence of fiber delay lines (FDL). In this setting, we investigate QoS differentiation via FDL access limitation. In other words, one class has access to the entire FDL bank whereas the other can access only a portion of the FDL bank. We show that there exists a trade-off between overall blocking probability and the level of class separation.
There is no registration fee, but registration is compulsory for organizational purposes.
Please follow this link to register online.
Link to Information Page
On the information page, also more about our preferred hotel, for those who need accommodation and need to book this personally.