Highlighted Work

Choose a topic from the list:

Software Defined Networks

As SDN technology is gradually deployed, the networks become softwarized and they can be optimized centrally by solving large instances of optimization problems. Traffic engineering algorithms are on the spotlight again, however with a new twist: to optimize globally softwarized networks we desire accurate and fast algorithms to solve large problems, but we have availability of computing resource.

In network slicing, the goal is to reserve resources including networking, computing, and storage, to provide survivable network abstractions to services with QoS guarantees. A major technological challenge is the resource management. In ‘‘The Algorithmic Aspects of Network Slicing’’ we describe an algorithmic framework for network slicing resource management problems.

We discuss key research challenges that relate to online algorithms and reactivity in ‘‘Online and Global Network Optimization: Towards the Next-Generation of Routing Platforms’’.

In ‘‘Controlling Flow Reconfigurations in SDN’’ we propose a new model for studying the flow reconfigurations in SDN. We define a control problem, which is to choose the time instances at which the output of the optimizer is actually applied to the real system, subject to a budget for reconfigurations. See also C.43.

C.36 VirtueMAN: A Software-Defined Network Architecture for WiFi-based Metropolitan Applications IEEE CAMAD 2014
We develop a NFV-based system that can be deployed on commodity WiFi hardware to turn the public devices into a Metropolitan Area Network (MAN).

Go Top


Caching has been proposed in networking as a means to reduce traffic and relief congestion. The majority of the volume of Internet traffic today is cacheable, meaning that it can be effectively stored and reproduced. By storing popular files near the edge of the network, we may avoid retransmitting the same file again and again, thereby reducing significantly the backbone network traffic. A key challenge is the fact that caches often are able to store only a tiny fraction of the entire file population. We discuss key research directions for wireless caching in Wireless Caching: Technical Misconceptions and Business Barriers.

J.11 Asymptotic Laws for Joint Content Replication and Delivery in Wireless Networks IEEE Trans. on Information Theory 2012
In this work we derive asymptotic laws for the required link capacity for the network to be sustainable. We assume that the network size and the file popularity grow to infinity. Ideally, we would like that the required link capacity does not scale (i.e. it is O(1)), but this is only true for particular popularity parameters. See also our INFOCOM paper C.22, and the extension to scaling cache size J.13.

C.41 Placing Dynamic Content in Caches with Small Population IEEE INFOCOM 2016
Most contents have time-varying popularity, while the majority of the literature assumes that content popularity is static. In this paper we develop the optimal joint learning and caching policy. Furthermore, we show that learning at a higher layer in the hierarchy is faster since more statistics can be aggregated. Last, we explain the tradeoffs when learning in clusters of geographical locations.

Report Adapting Caching to Audience Retention Rate: Which Video Chunk to Store?
Users often abandon the videos before they finish. In this work we provide a theoretical framework that analyzes the gains from applying partial caching to partially watched videos.

J.9 Storage Planning and Replica Assignment in Content-Centric Publish/Subscribe Networks Computer Networks 2011
In this work we study the content placement problem in order to minimize network traffic in the context of content-centric networking (CCN). This a hard combinatorial problem, therefore we propose a greedy heuristic algorithm. In C.12 we propose a caching framework for CCN, and in C.16 we study the user mobility aspects.

Go Top

Backpressure and Stability

Backpressure is dynamic routing algorithm that provides maximum throughput by balancing neighboring network queues (pressures), see this article.

C.42 Streaming Big Data meets Backpressure in Distributed Network Computation IEEE INFOCOM 2016
In this work we extend backpressure to include packet computations.

C.37 Throughput Optimal Broadcast for Wireless Directed Graphs IEEE INFOCOM 2015
In this work we apply backpressure to broadcast sessions. See also J.19 and C.46.

We propose two different ways of reducing the delay of backpressure:
1) C.38 Loop-Free Backpressure Routing Using Link-Reversal Algorithms ACM Mobihoc 2015
This approach uses a directed acyclic subgraph as biased for the packet routing to reduce loops. The subgraph is built dynamically according to queue lengths.

2) C.34 An Overlay Architecture for Throughput Optimal Multipath Routing ACM Mobihoc 2014
This approach applies backpressure only at a subset of nodes, while the rest perform shortest path routing. We give an algorithm to select the smallest set that guarantees the full multicommodity region. See also C.35, where we prove the optimality of a threshold-based policy controlling the tunnels.

J.10 Scheduling with pairwise XORing of packets under statistical overhearing information and feedback Queueing Systems 2012
When packets can have different states, it is possible to construct a virtual queueing network where each queue represents one state. For example in network coding, a packet may be decodable if combined with a subset of packets or codes which varies over time (e.g., due to successive transmissions). Hence over time the same packet may traverse several decodability states. In this paper we show that backpressure can be applied on the virtual network to yield a throughput optimal policy for the original problem. In C.27 we successfully apply this tool to the case of coded opportunistic routing.

J.14 Minimal Evacuation Times and Stability IEEE/ACM Trans. on Networking 2014
The analysis systematizes the use of the minimum evacuation time of packets in a network. Our work proves that the evacuation function characterizes completely the system throughput. Moreover, evacuating packets in batches allows development of optimal control policies. This result facilitates the analysis of systems with tractable evacuation performance, examples of which include broadcast channels C.24, wireless network coding C.25, C.31, and input queued switches J.14.

Go Top

Datacenters and Clouds

C.44 Routing with Blinkers: Online Throughput Maximization without Queue Length Information ISIT 2016
We propose a distributed load balancing scheme where each load balancer decides routing based on the delays experienced by past routed jobs from itself. We give the conditions under which this scheme is optimal.

C.33 Dynamic Overload Balancing in Server Farms IFIP Networking 2014
Datacenters may experience situations of overload. This work proposes an optimization framework for smoothly handling situations of overloading.

J.16 CPU Provisioning Algorithms for Service Differentiation in Cloud-based Environments IEEE Trans. on Network and Service Management 2015
Further, we consider the case of splitting CPU rates to servers based on QoS classes.

J.18 Sustainability of Service Provisioning Systems under Stealth DoS Attacks IEEE TCNS 2016
We consider datacenters attacked by botnets. We study the complex interaction of two schedulers, the defending trying to balance the load on the servers, and the attacking trying to destabilize it.

Go Top

Congestion Control

To optimally resolve network congestion, we employ the Network Utility Maximization framework. The utility of network users is expressed as a concave function of user throughput, and the employed policies decide how to inject packets in the network and how to route them, so that the time-average utility is maximized.

In a line of works we consider the case of controlling the injected packets implicitly from the receivers. The receivers create pressure and the sources consider the propagated pressure to inject new packets or not.

J.17 In-network Congestion Control for Multirate Multicast Accepted in IEEE/ACM Trans. on Networking 2016
This paper derives the optimal receiver-based congestion control for multi rate multicast sessions. Multi-rate multicast is a scheme used to deliver several levels of video quality to users. See also our INFOCOM paper C.32.

C.39 Distributed Load Shedding with Minimum Energy IEEE INFOCOM 2016
We further consider the case of applying congestion control inside the network. Each network node can decide how many of the arriving packets should be forwarded to the next hop, and how many should be dropped.

Go Top

Network Coding for Wireless Systems

Network coding is a special type of code that occurs inside the network. It is known that network coding increases network throughput in some cases (in case of multicast, multiple commodities, broadcast with erasures, etc). See this article.

J.12 A Modular Framework for Implementing Joint Wireless Network Coding and Scheduling Algorithms EURASIP JWCN 2013
We develop the NCRAWL scheme: Network Coding scheme with Rate Adaptive Wireless Links. NCRAWL extends the well-known COPE scheme to exploit variable channel rates and congestion feedback in order to increase throughput.

J.15 Dynamic Wireless Network Coding with Overhearing and Variable Channel Rates IEEE JSAC 2015
See also our INFOCOM paper C.33, and C.31, where we make a connection of wireless network coding with overhearing to index coding. In C.27 we use this methodology to find the maximum throughput of opportunistic routing.

J.10 Scheduling with pairwise XORing of packets under statistical overhearing information and feedback Queueing Systems 2012
We study scheduling and routing of packets jointly with network coding. We derive the optimal policy under the restricted case of XOR coding.

Go Top

Energy Saving in Wireless

Saving energy in communication networks is very important, and mainly focused on (i) saving battery of mobile terminals, (ii) reducing monthly costs of networks, (iii) reducing the CO_2 footprint of ICT.

C.21 Medium access games: the impact of energy constraints NetGCoop 2011
In this study, we look at ALOHA-type randomized protocols, and consider each user as a player in a game. The goal of the player is to maximize throughput, where the energy is naturally constrained by available battery life. We prove that the energy constraints lead the game to an equilibrium with bounded price of anarchy. This is contrary to classical ALOHA games which are known to have unbounded price of anarchy.

J.7 Optimal control of sleep periods for wireless terminals IEEE JSAC 2011
We focus on ‘‘sleep modes’’, i.e., the mobiles turn off the transceiver and go to an ambient mode in order to save energy. The question is how often should the mobile wake up to poll the base station for waiting packets. WIMAX standards define a increasing window protocol. This work establishes the optimal sleep modes for several traffic models, exponential (Poisson arrival) and hyper-exponential (used to model self-similar traffic and packet bursts).

C.13 Providing quality of service guarantees in multiclass IEEE 802.16e sleep mode IEEE GLOBECOM 2009
In this work we derive analytical formulas for the energy saving of the WIMAX sleep mode protocol with multiple classes.

Go Top

Wireless LANs

WiFi networks are built according to the IEEE 802.11 standard, and utilize randomized transmissions. In this line of work we evaluate the performance of such networks and propose techniques that improve their performance by tweaking different parameters.

J.3 A new MAC protocol for supporting Quality of Service in 802.11 Wireless LANs EURASIP JWCN 2006
This work proposes the use of admission control and time synchronization to distributedly form a TDMA-like frame for WiFi devices. The system is backward compatible and outperforms classical random access. It has been implemented in this recent work.

C.8 Extension and Comparison of QoS-Enabled Wi-Fi models in the presence of errors IEEE GLOBECOM 2007
We analyze IEEE 802.11e in the presence of channel errors. This work received the best paper award.

Go Top