Vitor Jesus

Mobility with QoS Support for Multi-Interface Terminals: Combined User and Network Approach

In the future, mobile terminals will be equipped with several interfaces to different technologies -such as UMTS, IEEE 802.11, DVB and WiMAX -and will be able to connect simultaneously to the most appropriate access networks. This paper presents a mobility and QoS architecture aiming at supporting intelligent handovers. We introduce the concept of Network-Assisted Mobile terminal Initiated HandOver (NAMIHO), where the handover decision is negotiated between the network and the user, taking into account inputs and decision algorithms running both on the terminal and network. We also present an instantiation of the architecture using IEEE 802.21 and concepts of local/global mobility. We further propose extensions to IEEE 802.21 standard.

Any-constraint personalized network selection

In densely covered areas and with multi-interface terminals, network selection is a complex problem, especially if it takes into account different criteria such as context and signal strength. We present an algorithm that allows a mobility decision manager to consider arbitrary criteria such as user history or preferences. Since a typical mobility module uses quantitative and well-known information (eg., radio signal strength or available resources), qualitative information must be fed into the mobility subsystem only after converting it to a suitable format. Our algorithm uses (up to) three stages to produce a decision: first, it discards unsuitable possibilities; second, it produces a set of possibilities considering resource availability; and third, the feasible possibilities are ranked according to contextual information.

Virtual Network Capacity Expansion through Service Outsourcing

Roaming agreements in 3G and beyond 3G networks can greatly enhance the delivery of services to end users. The cost of service delivery to the user depends on the price charged on networks traversed by user traffic. In a QoS DiffServ environment, network resources may be sold in aggregate blocks at wholesale prices to competing ISPs. This can influence the cost of delivering services between two points using access networks owned by competing operators. Demand for network services by users varies over time and network capacities are finite; thus a fully loaded network would reject new service requests and an underutilized network will become less productive. In this paper we explore a service outsourcing scheme between competing operators that allows a custodian network operating at full capacity to outsource service provision to a candidate network. Outsourcing would be price-influenced enabling the home (custodian) operator to levy local predictable charges to the end users for services offered on the visited network. This will virtually expand the capacity of the custodian network and boost incomes for both operators.

Figures of merit for the placement (in) efficiency of interconnected CDNs

CDN interconnection is currently being debated across industry, academia and standards organizations, both in terms of proposing mechanisms for inter-CDN cooperation, such as architectures and protocols, and in terms of the conditions under which cooperation is advantageous. In this paper we focus on the second part and provide contributions to the fundamental problem of when two CDNs should cooperate. Given two CDNs, we aim at quantifying the loss in provisioning efficiency that arises from inter-CDN cooperation. We solve a CDN placement problem in interconnection scenarios, in order to compare the cost of centrally provisioning a network with the joint cost of provisioning the same topology but partitioned and with both sub-topologies independently and only locally optimally provisioned. Given the complexity of the topic (particularly the computational cost) but also aiming at general figures of merit, we use a non-capacitated K-median problem and solve the problem for a large number of topologies. Among other results, we show that the incurred inefficiency of provisioning the same topology partitioned in two independent CDNs, but locally optimal, can raise up to 30%. Overall, we demonstrate with exact solutions that interconnection may not be straightforwardly advantageous.

Linking interdomain business models to the current Internet topology

The Internet is built on a model in which Autonomous Systems (AS) peer bilaterally based on pair-wise Service Level Agreements (SLA). This results in a cascaded connectivity model in which routing decisions beyond the next hop are delegated, so an AS has limited control over paths beyond the next hop. While this is a key feature of the Internet, the impact of this cascading model on the structure of the Internet has, to the best of our knowledge, never been analyzed in detail. Using game theory with realistic rules, we provide an analysis of the cascading model. Although our model does not intend to capture the full economic and technical complexity of the real Internet, we show that it exposes the two main business models of peering used currently: volume-based peering and transit carriers (large hubs with equivalent capacity).

A stateless architectural approach to inter-domain QoS

QoS provisioning as a generic telecom service is only useful if deployed end-to-end, so it must cover both the access segment and the interdomain segment of routes. This work focuses on the interdomain component of QoS architectures. Although the research community has been discussing several aspects of the problem, integrated architectures for true end-to-end QoS are still largely conceptual, especially if one considers mesh models and not simpler cascaded ones. Our inter-domain proposal combines offline service discovery and registration with real-time congestion notification, allowing us to significantly reduce the number of packets that were not suitably serviced, in the sense of fulfilling required Service Level Agreement constraints. We qualify our approach of dasiastatelesspsila since our proposal doesnpsilat use additional state in forwarding elements which is one of the most dramatic problems for interdomain scenarios.

Topological Implications of Cascading Interdomain Bilateral Traffic Agreements

The Internet uses a model in which Autonomous Systems (AS) peer bilaterally with each other, resulting in a cascaded connectivity model: the end-to-end service that an end-user sees is the result of this cascade of bilateral traffic agreements. As a result, the routing decisions at each AS beyond the next hop are implicitly delegated, so an AS has limited control over the remaining path. While this may be considered a key feature of the Internet (e.g., helps scalability), the impact of this cascading model on the structure of the Internet is not yet well understood. In this paper, we analyze this cascaded model using concepts of game theory. Although our model cannot capture the full complexity of the real Internet - we actually aim at simplicity and try to only isolate the cascading effect -, our results suggest that cascading bilateral agreements brings order to an arbitrary graph. In particular, it brings forward some well-known properties of the Internet topology, such as the current AS hierarchy and common peering models. Finally, we compare our results with topological and routing data, obtained from CAIDA, looking for experimental evidence of our results, while further exploring the insight obtained from our model.

Discovery and Composition of Per-Domain Behaviours - a Service Abstraction Approach

We discuss the problem of composing Quality-of-Service across several administrative domains considering an evolved DiffServ notion of Per-Domain Behavior (PDB). We discuss two components of the problem: path discovery and the effect of composing PDBs. We assume the most general scenario: domains are free to adopt any PDB, as long there are common semantics and a set of well-known QoS parameters (e.g. one-way delay, peak/sustained bandwidth, etc.). Thus we adopt an abstract representation of PDBs. We obtain three main results. First, we show that a distributed path discovery scheme is feasible and more scalable than a centralized one. Second, we show the outcome of inter-domain QoS composition, for general metrics and for Internet-alike topologies. Finally, and as an important practical result, we show that the Internet may not need a centralized governance model, in terms of the definition of inter-domain PDBs.

Supporting Dynamic Inter-Domain Network Composition: Domain Discovery

When two administratively independent domains need to engage in cooperation of some kind, some initial common known point of contact must exist, both in terms of the application topological position (i.e., IP address, transport protocol and port) and in terms of the interface technology (e.g., the protocol suite of the interaction). In a world of many domains of many types (not necessarily Autonomous Systems in the sense of interdomain routing), manual or single-directory-based operations are not practical (or even feasible); hence, some form of autonomic discovery and handshaking of domains is needed. We propose and evaluate several strategies to allow autonomic bootstrapping of inter-domain operations, all fit to be deployed as BGP extensions. We show the time/message overhead tradeoff: its possible to obtain the absolute minimum message complexity of O(N) but at the cost of central administrative burden or, for fully distributed schemes, a tradeoff (message/time complexity) of O(N2)/O(N) or O(N2)/O(log2N).

A simulation study on the wired-equivalent capacity of ad hoc networks

In this paper we demonstrate the concept of wired- equivalent capacity in ad-hoc networks through a large-scale simulation study. A capacity model that mimics conventional experience (e.g. on wired, infrastructured networks) is proposed under a specific, yet realistic and user-oriented, assumption of the maximum fair usage. Capacity is then the maximum throughput capacity per node under which packet loss is similar to that of a conventional wired network, for all nodes. We show how this capacity metric becomes predictable inside ad-hoc networks using a simple power-law while delay is bounded.