Multi-user video conferencing is a ubiquitous technology. Increasingly end-hosts in a conference are assisted by cloud-based servers that improve the quality of experience for end users. This paper evaluates the impact of strategies for placement of such servers on user experience and deployment cost. We consider scenarios based upon the Amazon EC2 infrastructure as well as future scenarios in which cloud instances can be located at a larger number of possible sites across the planet. We compare a number of possible strategies for choosing which cloud locations should host services and how traffic should route through them. Our study is driven by real data to create demand scenarios with realistic geographical user distributions and diurnal behaviour. We conclude that on the EC2 infrastructure a well chosen static selection of servers performs well but as more cloud locations are available a dynamic choice of servers becomes important.

Cloud computing has revolutionized service delivery by providing on-demand invocation and elasticity. To reap these benefits, computation has been displaced from client devices and into data centers. This partial centralization is undesirable for applications that have stringent locality requirements, e.g., low latency. This problem could be addressed with large numbers of smaller cloud resources closer to users. However, as cloud computing diffuses from within data centers and into the network, there will be a need for cloud resource allocation algorithms that operate on resource-constrained computational units that serve localized subsets of customers. In this paper, we present a mechanism for service provisioning in distributed clouds where applications compete for resources. The mechanism operates by enabling execution zones to assign resources based on Vickrey auctions and provides high-quality probabilistic models that applications can use to predict the outcomes of such auctions. This allows applications to use knowledge of the locality distribution of their clients to accurately select the number of bids to be sent to each execution zone and their value. The proposed mechanism is highly scalable, efficient, and validated by extensive simulations.

TCP remains the dominant transport protocol for Internet traffic, but the preponderance of its congestion control mechanisms in determining flow throughput is often disputed. This paper analyzes the extent to which network, host and application settings define flow throughput over time and across autonomous systems. Drawing from a longitudinal study spanning five years of passive traces collected from a single transit link, our results show that continuing OS upgrades have reduced the influence of host limitations owing both to windowscale deployment, which by 2011 covered 80% of inbound traffic, and increased socket buffer sizes. On the other hand, we show that for this data set, approximately half of all inbound traffic remains throttled by constraints beyond network capacity, challenging the traditional model of congestion control in TCP traffic as governed primarily by loss and delay.

Existing methods for traffic resilience at the network and transport layers typically work in isolation, often resorting to inference in fault detection and recovery respectively. This both duplicates functionality across layers, eroding efficiency, and leads to protracted recovery cycles, affecting responsiveness. Such misalignment is particularly at odds with the unprecedented concentration of traffic in data-centers, in which network and hosts are managed in unison. This paper advocates instead a cross-layer approach to traffic resilience. The proposed architecture, INFLEX, builds on the abstractions provided by software-defined networking (SDN) to maintain multiple virtual forwarding planes which the network assigns to flows. In case of path failure, transport protocols pro-actively request to switch plane in a manner which is unilaterally deployable by an edge domain, providing scalable end-to-end forwarding path resilience.

This paper describes TARDIS (Traffic Assignment and Retiming Dynamics with Inherent Stability) which is an algorithmic procedure designed to reallocate traffic within Internet Service Provider (ISP) networks. Recent work has investigated the idea of shifting traffic in time (from peak to off-peak) or in space (by using different links). This work gives a unified scheme for both time and space shifting to reduce costs. Particular attention is given to the commonly used 95th percentile pricing scheme.

Management operations performed by Content Delivery Network (CDN) providers consist mainly in controlling the placement of contents at different storage locations and deciding where to serve client requests from. Configuration decisions are usually taken by using only limited information about the carrier networks, and this can adversely affect network usage. In this work we propose an approach by which ISPs can have more control over their resources. This involves the deployment of caching points within their network, which can allow them to implement their own content placement strategies. The work presented in this paper investigates lightweight strategies that can be used by the ISPs to manage the placement of contents in the various network caching locations according to user demand characteristics. The proposed strategies differ in terms of the volume and nature of the information required to determine the new caching configurations. We evaluate the performance of the proposed strategies, in terms of network resource utilization, based on a wide range of user demand profiles and we compare the obtained performance according to metrics we define to characterize the demand. The results demonstrate that the proposed metrics can provide useful indications regarding the performance one strategy can achieve over another and, as such, can be used by the ISP to improve the utilization of network resources.

This paper considers fundamental measurements which drive TCP flows: throughput, RTT and loss. It is clear that throughput is, in some sense, a function of both RTT and loss. In their seminal paper Padyhe et al [1] begin with a mathematical model of the TCP sliding window evolution process and come up with an equation showing that TCP throughput is (roughly) proportional to 1/RTT√p where p is the probability of packet loss. Their equation is shown to be consistent with data gathered on several links. This paper takes the opposite approach and analyses a large number of packet traces from well-known sources in order to create a data-driven estimate of the functions which relate TCP, loss and RTT. Regression analysis is used to fit models to connect the quantities. The fitted models show different behaviour from that expected in [1].

When designing distributed systems and Internet protocols, designers can benefit from statistical models of the Internet that can be used to estimate their performance. However, it is frequently impossible for these models to include every property of interest. In these cases, model builders have to select a reduced subset of network properties, and the rest will have to be estimated from those available. In this paper we present a technique for the analysis of Internet round trip times (RTT) and its relationship with other geographic and network properties. This technique is applied on a novel dataset comprising ~19 million RTT measurements derived from ~200 million RTT samples between ~54 thousand DNS servers. Our main contribution is an information-theoretical analysis that allows us to determine the amount of information that a given subset of geographic or network variables (such as RTT or great circle distance between geolocated hosts) gives about other variables of interest. We then provide bounds on the error that can be expected when using statistical estimators for the variables of interest based on subsets of other variables.

There is growing evidence that mutually beneficial outcomes can be achieved when content distribution overlays and their underlying ISPs collaborate through open interfaces. We further contribute to this body of work by considering consolidated topology construction strategies that integrate the information provided by multiple ISPs. We focus on situations with potentially conflicting, asymmetric preference costs, since these situations are expected to benefit more from the tradeoffs provided by consolidation to produce an overlay topology with desirable global properties. In this paper we develop a generic model for the multi-domain consolidation of ISP preferences expressed as costs for pairwise peer connections, where peers are grouped into clusters based on topology criteria. Using this model, we propose two consolidated topology construction strategies: Shared Cost, designed to provide a tradeoff for preference cost asymmetries, and Low Cost, designed to reduce the overall preference cost that the overlay imposes on all its underlying ISPs. We evaluate these two models through extensive simulations over a wide range of ISP and peer cluster sizes, and we show that preference consolidation can provide ISPs with outcomes more aligned with their preferences than those provided by non-consolidated operation.

The popularity of video sharing over the Internet has increased significantly. High traffic generated by such applications at the source can be better distributed using a peer-to-peer (P2P) overlay. Unlike most P2P systems, LiveShift combines both live and on-demand video streaming while video is transmitted through the peer-to-peer network in a live fashion, all peers participate in distributed storage. This adds the ability to replay time-shifted streams from other peers in a distributed and scalable manner. This paper describes an adaptive fully-distributed mesh-pull protocol that supports the envisioned use case and a set of policies that enable efficient usage of resources, discussing interesting trade-offs encountered. User-focused evaluation results, including both channel switching and time shifting behavior, show that the proposed system provides good quality of experience for most users, in terms of infrequent stalling, low playback lag, and a small proportion of skipped blocks in all the scenarios studied, even in presence of churn.

This paper presents Packet Re-cycling (PR), a technique that takes advantage of cellular graph embeddings to reroute packets that would otherwise be dropped in case of link or node failures. The technique employs only one bit in the packet header to cover any single link failures, and in the order of log2(d) bits to cover all non-disconnecting failure combinations, where d is the diameter of the network. We show that our routing strategy is effective and that its path length stretch is acceptable for realistic topologies. The packet header overhead incurred by PR is very small, and the extra memory and packet processing time required to implement it at each router are insignificant. This makes PR suitable for loss-sensitive, mission-critical network applications.

In this paper, we propose a routing technique, “recursive Loop-Free Alternates (rLFAs)”, to alleviate packet loss due to transient link failures. The technique consists of a backup path calculation with corresponding re-routing scheme based on the Loop-Free Condition (LFC) as defined in the basic specification for IP Fast Re-Route (IPFRR). Under this routing strategy, nodes calculate backup paths by modifying the weights of links in the primary shortest path tree. If a failure occurs, the detecting node determines the number of recursions, which indicates the number of times packets must be forwarded along the alternate next hops to bypass the failed link. This technique guarantees full repair coverage for single link failures. We evaluate the performance of our proposed technique through simulations and show that the incurred overheads, the stretch of its pre-computed alternate paths, and the failure-state Maximum Link Utilisation (MLU) are minimal.

This paper criticises the notion that long-range dependence is an important contributor to the queuing behaviour of real Internet traffic. The idea is questioned in two different ways. Firstly, a class of models used to simulate Internet traffic is shown to have important theoretical flaws. It is shown that this behaviour is inconsistent with the behaviour of real traffic traces. Secondly, the notion that long-range correlations significantly affects the queuing performance of traffic is investigated by destroying those correlations in real traffic traces (by reordering). It is shown that the longer ranges of correlations are not important except in one case with an extremely high load.

Although direct reciprocity (Tit-for-Tai) contribution systems have been successful in reducing freeloading in peer- to-peer overlays, it has been shown that, unless the contribution network is dense, they tend to be slow (or may even fail) to converge. On the other hand, current indirect reciprocity mechanisms based on reputation systems tend to be susceptible to sybil attacks, peer slander and whitewashing. In this paper we present PledgeRoute, an accounting mechanism for peer contributions that is based on social capital. This mechanism allows peers to contribute resources to one set of peers and use this contribution to obtain services from a different set of peers, at a different time. PledgeRoute is completely decentralised, can be implemented in both structured and unstructured peer-to-peer systems, and it is resistant to the three kinds of attacks mentioned above. To achieve this, we model contribution transitivity as a routing problem in the contribution network of the peer-to-peer overlay, and we present arguments for the routing behaviour and the sybilproofness of our contribution transfer procedures on this basis. Additionally, we present mechanisms for the seeding of the contribution network, and a combination of incentive mechanisms and reciprocation policies that motivate peers to adhere to the protocol and maximise their service contributions to the overlay.

The production of a large-scale monitoring system for a high-speed network leads to a number of challenges. These challenges are not purely technical but also socio-political and legal. The number of stakeholders in such monitoring activity is large including the network operators, the users, the equipment manufacturers and, of course, the monitoring researchers. The MASTS project (measurement at all scales in time and space) was created to instrument the high-speed JANET Lightpath network and has been extended to incorporate other paths supported by JANET(UK). Challenges the project has faced included: simple access to the network; legal issues involved in the storage and dissemination of the captured information, which may be personal; the volume of data captured and the rate at which these data appear at store. To this end, the MASTS system will have established four monitoring points each capturing packets on a high-speed link. Traffic header data will be continuously collected, anonymised, indexed, stored and made available to the research community. A legal framework for the capture and storage of network measurement data has been developed which allows the anonymised IP traces to be used for research purposes.

Peer-to-peer networks providing QoS-enabled services are sensitive to hidden action situations, where the actions of a server peer are hidden from the peers who receive services from it. This is because server peers can choose to strategically minimize their effort, and client peers may be unable to distinguish between cases where the server exerted insufficient effort and cases where the server kept its advertised effort levels but the end-to-end conditions in the network were sufficiently adverse. We propose a principal-agent model for hidden action that gives server peers sufficient incentives to meet their advertised effort levels, without client peers having to decide for each transaction whether the outcome was due to server behavior or network conditions. This allows peers to draft contracts that provide incentives for truthful revelation of QoS capabilities, and to have predictable transaction quality. We then exemplify the model for the case of a mesh-based, pull-oriented streaming system with low delay requirements. For this case, we show how to estimate the model parameters a function of the prevailing network conditions, and how to enforce contract fulfillment through a reciprocative strategy.