Research projects at LCN
We conduct research on most aspects of networking, in particular network and service management, router architectures, wireless networks for delay tolerant broadcast and for video communication, peer-to-peer networking and sensor network dimensioning and routing. We are seeking designs for low complexity, good scalability, ease of analysis and conceptual simplicity. Our research is based on a blend of analysis, simulation and experimentation. We have a extensive research laboratory with state-of-the-art equipment.
Adaptive decentralized monitoring
We engineer a set of protocols that execute in a in a distributed management layer, performing the tasks of distributed polling, continuous estimation, and threshold detection for network-wide aggregates. In order to keep the complexity of the management layer low and to enable efficient, effective and scalable operation, these protocols are self-configuring, robust, and tunable at runtime. This work is performed in cooperation with Cisco Systems, USA.
People involved: Rolf Stadler, Alberto Gonzales, Fetahi Wuhib
Self-organizing server clusters
We develop a peer-to-peer service management middleware that dynamically allocates system resources to a large set of applications. The system achieves scalability in number of nodes (1000s or more) through decentralized mechanisms that run on different time scales. It adapts to changes in load pattern, system failures, and management policies. This is performed in cooperation with IBM Research, USA.
People involved: Rolf Stadler, Constantin Adam
Auto-configuration and self-healing of radio access networks
We investigate techniques for decentralized configuration of neighboring cell lists. Specifically, we study the use of Bloom filters and in-network aggregation techniques to achieve low overhead and fast reaction times. This work is performed in cooperation with Ericsson Research.
People involved: Rolf Stadler, Javier Baliosan
Peer-to-peer media streaming
Peer-to-peer is a promising method for large scale content distribution, such as IPTV and video on demand, since it overcomes many limitations of the traditional client-server paradigm. P2P streaming faces however several challenges: The number of peers participating in the overlay may change rapidly; the streams have to be transmitted with end-to-end delays acceptable for live applications, which are in the range of a couple of tenths of seconds; in order to keep the perceived media quality acceptable, the packet loss rate has to be low. We define mathematical models that describe the data distribution process in P2P overlays. The models allow us to compare various architectures and traffic control solutions with respect to their efficiency, and to define new methods to improve the streaming performance. Our goal is to define the building blocks that are necessary and sufficient to provide large scale streaming with good perceived quality. This research is financed by a Wireless at KTH small project grant, and by the SSF Winternet and AWSI projects.
People involved: Viktoria Fodor, György Dán, Ilias Chatzidrossos
We design an open, receiver-driven broadcasting system that relies on delay-tolerant forwarding of data chunks through the mobility of wireless nodes. The system provides public broadcast channels, which are openly used for both transmission and reception. We show by analysis and simulation that a delay-tolerant broadcast channel has both a sufficiently high throughput and reach to be interesting as a competitive alternative to the regulated wireless broadcast channel. This work is performed in collaboration with Vincent Lenders and Martin May at ETH Zurich. The work is supported by the SSF Winternet grant.
People involved: Gunnar Karlsson, Vladimir Vukadinovic, Olafur Ragnar Helgason
Wireless video communication
One of the central problems in video transmission over lossy time-varying channels is to allocate optimally the available resources, such as the bit-rate and the right to access the channel. We study both different strategies for choosing source and channel coding rates and the ability of opportunistic scheduling to meet latency requirements of streaming applications. Our study also considers resource sharing for integration of multicast and unicast data on time-slotted radio channels. Our results show that a simple bit-rate allocation strategy for minimizing instantaneous distortion is not necessarily optimal in terms of time-average distortion and that opportunistic scheduling faces major difficulties to ensure good user-level performance when streaming flows constitute a significant share of a traffic load. The performance gains of resource-optimal multicast could be substantial and should be exploited in practical scheduling schemes. The work is supported by the SSF Winternet grant.
People involved: Gunnar Karlsson, Vladimir Vukadinovic
Centralized Control in IP Networks
We study a centralized control scheme for current IP networks, where the intra-domain routing is managed through centralized servers instead of decentralized link-state routing protocols. Centralized servers determine the forwarding paths for all routers based on the network topology and the traffic demands, and distribute the forwarding information to the routers. We consider several topics in the centralized control scheme. First, the convergence processes of both the centralized and the decentralized control schemes are studied. Second, we specify an order of routers to update their forwarding tables to avoid transient loops. Third, we study traffic engineering in networks with the centralized control scheme, where instead of manipulating traditional OSPF link weights, the forwarding tables can be modified directly. The work is supported by the SSF Winternet grant.
People involved: Gunnar Karlsson, Jing Fu
Dimensioning and routing for sensor networks
We study the connectivity pattern of wireless sensor networks where sensor nodes are scattered randomly to monitor an area and they need to communicate with a single sink node. In this scenario the sensor nodes may form clusters, where nodes within a cluster are connected, but are not connected to other nodes in the network. In this scenario, the placement of the sensor nodes can not be directly controlled; the only design parameter is the number of sensors scattered in the area. To capture this design parameter, we propose a random grid model to describe the scattering of the sensors. Based on the model, we evaluate how the size of the largest cluster changes with the density of the sensors, and what the required sensor density is to provide a well-connected sensor network while also considering node failures. In the next step of our research we will use the random grid model to propose routing algorithms for randomly scattered sensor networks and evaluate their performance in terms of energy efficiency, throughput and delay. This work is financed by the Nordite CROPS project.
People involved: Gunnar Karlsson, Viktoria Fodor, Alonso Zuniga