The peer-to-peer paradigm shows the potential to provide the same functionality like client/server based systems, but with much lower costs.

One substantial limitation of p2p systems are missing guarantees on the quality of service, as the whole infrastructure is based on unreliable peers. In order to control the quality of peer-to-peer systems, monitoring and management mechanisms need to be applied. Both tasks are challenging in large-scale networks with autonomous, unreliable nodes.

A substantial amount of work has recently gone into localizing BitTorrent traffic within an ISP in order to avoid excessive and often times unnecessary transit costs. Several architectures and systems have been proposed and the initial results from specific ISPs and a few torrents have been encouraging. In this work we attempt to deepen and scale our understanding of locality and its potential. Looking at specific ISPs, we consider tens of thousands of concurrent torrents, and thus capture ISP-wide implications that cannot be appreciated by looking at only a handful of torrents. Secondly, we go beyond individual case studies and present results for the top 100 ISPs in terms of number of users represented in our dataset of up to 40K torrents involving more than 3.9M concurrent peers and more than 20M in the course of a day spread in 11K ASes. We develop scalable methodologies hat permit us to process this huge dataset and answer questions such as: What is the minimum and the maximum transit traffic reduction across hundreds of ISPs? What are the win-win boundaries for ISPs and their users? What is the maximum amount of transit traffic that can be localized without requiring fine-grained control of inter-AS overlay connections? What is the impact to transit traffic from upgrades of residential broadband speeds?

This talk addresses the problem of providing throughput guarantees in heterogeneous wireless mesh networks. As a first step, it proposes a novel model to represent the capacity region of a wireless link that, by linearizing this region, has the fundamental property of being very simple while providing a good approximation to the entire region. In a second step, this model is mapped to two of the most prominent wireless technologies nowadays, namely Wireless LAN and WiMAX. The last step addresses the issue of finding optimal routing strategies, which is done by solving an optimization problem subject to the constraints imposed by the linearized capacity region. The performance of the proposed approach has been compared against traditional routing metrics in mesh networks, such as ETT and ETX, and shown to overperform them by approximately a factor of 2.

The way the Internet works today is that routing protocols select a single path through the network, and transport protocols like TCP then adapt to the properties of that path. So the situation where there are unused paths that remain idle is quite common. The idea behind multipath TCP is to modify TCP so it can send packets over multiple paths at the same time, moving traffic from more congested paths to less congested paths and thus optimizing network utilization and performance.

Mobility is, without doubt, one of the major new paradigms of the current Internet, and this is driving most of the research activity in networking throughout the World. We are gradually evolving from a network where most end-systems have a fixed or quasi-fixed point of attachment, towards a global network where end systems seamlessly move from network to network and where networks themselves change their connection point to the Internet. In spite of this there is still a long way to go before user and network mobility – at IP level – become a reality, as several technological and research challenges persist.

Many delay-based congestion protocols have been proposed. Some recent studies have questioned the validity of congestion prediction at end hosts Based on measurement studies. In this talk, we show that end-host based delay prediction can be more accurate than previously characterized. We propose PERT (Probabilistic Early Response TCP) to mitigate the uncertainties in end-host based congestion prediction. PERT emulates the behavior of AQM/ECN in the end hosts' response to congestion.

The problem is interesting for two reasons. First, the cost function closely models the energy consumption of some network elements and network-wide optimization is a well-motivated but under-explored direction for energy minimization. Second, it brings light to a challenging combinatorial optimization problem. We will present positive and negative results for polynomial functions and polynomial functions with startup cost. For the latter, techniques to accomplish better-than-polynomial approximation ratios independent of demands and cost function remains a challenging problem.

Energy consumption in Computer Networks is an issue that is gaining increasing attention. In this talk we will review the issue of energy consumption in Computer Networks to then focus on Ethernet where a new standard is being developed to improve energy efficiency. We will discuss some of the alternatives that were considered during the Energy Efficient Ethernet standardization process and describe the solution adopted in the standard. Then we will present some results of an initial performance evaluation of the standard that shows the potential savings for different scenarios. We will conclude the talk by outlining the implications that Energy Efficient Ethernet will have on different aspects of the LANs.

The Theory of Computing is almost a hundred year old now. Its roots can actually be traced to the Entscheidungsproblem posed by David Hilbert in 1928. Another concrete theory in Physics called the Quantum Mechanics had its inceptions almost two hundred years ago (with Thomas Young's Double Slit Experiment in 1803) but actually started in the late 19th century. Today, Quantum Mechanics is proven to be the most successful theory of Physics. So the natural question to ask was that while the statement of Church-Turing Thesis is seen to be a statement of physics (laws of nature), then it should be compatible with the theory of Quantum Mechanics!

Existing cloud computing platforms offer virtually unlimited compute resources (virtual machines, bandwidth, storage, etc.) that can be used on demand. Such on-demand model offers significant elasticity to the customers in terms when and where they use the resources. The existing pricing model, however, is pay-as-you-go which in turn can lead to unpredictable costs to the cloud customers. This talk will discuss two adaptive approaches for resource control under a fixed budget: Distributed Rate Limiting (DRL) and Temporal Rate Limiting (TRL). DRL is a fully decentralized mechanism for resource control over a distributed cloud service, that splits the available budget among the participating nodes subject to the load each node experiences. TRL in contrast, splits the budget over a time period, to optimize the performance of the customer with demand pattern that varies in time.