Hélia Pouyllau

End-to-end QoS negotiation in network federations

The Future Internet will see the emergence of value-added applications (e.g. telepresence, etc.) requiring Quality of Service (QoS) guarantees across multiple managed networks. Deploying mechanisms to solve technical questions entails a minimal cooperation between network operators (aka. carriers). But, this cooperation is subject to economic constraints (e.g. confidentiality on resources, pricing, risk avoidance, etc.). To overcome them, an interesting direction lies in the creation of operator federations, aka. alliances, that would ensure a technical convergence. In this paper, we detail the work in progress on a scenario of inter-carrier service negotiation taking place within such a federation. The service is negotiated between a third party responsible for the federation and customers. We describe the associated game and analyze equilibrium conditions in it.

Inter-carrier SLA negotiation using Q-learning

Inter-domain high performance services (e.g. telepresence) are not sustainable over the current Internet architecture. The Quality of Service (QoS) guarantees they demand require to settle on end-to-end Service Level Agreements (SLAs) among providers (aka. carriers) and across different networks. This process is critical since it must provide the most benefits while dealing with heterogeneous operators’ business interests and confidentiality constraints. In this paper, we propose, in the frame of a cooperative organizational model called federation, a composition technique for inter-carrier SLAs that respects end-user’s QoS requirements while maximizing network operators’ long-term benefits. We formulate the dynamic optimization problem as a Markov Decision process (MDP). This latter allows to provide an iterative near-optimal solution through reinforcement learning (more precisely, Q-learning). The SLA composition is thus performed taking into account customers and network providers’ utilities. We also propose a version including several negotiation rounds and observe how it affects the results.

Distributed inter-domain SLA negotiation using Reinforcement Learning

1 Applications requiring network Quality of Service (QoS) (e.g. telepresence, cloud computing, etc.) are becoming mainstream. To support their deployment, network operators must automatically negotiate end-to-end QoS contracts (aka. Service Level Agreements, SLAs) and configure their networks accordingly. Other crucial needs must be considered: QoS should provide incentives to network operators, and confidentiality on topologies, resource states and committed SLAs must be respected. To meet these requirements, we propose two distributed learning algorithms that will allow network operators to negotiate end-to-end SLAs and optimize revenues for several demands while treating requests in real-time: one algorithm minimizes the cooperation between providers while the other demands to exchange more information. Experiment results exhibit that the second algorithm satisfies better customers and providers while having worse runtime performances.

Distributed Optimization Algorithms for X-Domain End-to-End QoS Negotiation

Establishing end-to-end services in X-domain networks is an untrivial problem since it implies to guarantee end-to-end Quality of Service (QoS). The contract approach consisting in designing local QoS contracts (Service Level Agreements, SLAs) seems to be the most adapted approach to achieve this goal. However, due to the independent and heterogenous nature of the Internet domains, the problem may be solved only through a distributed process respecting also the contract privacy. In previous works, we described some distributed optimization algorithms to determine such contract chains. In this paper, we propose an improvement of one of our previous solution. We also give the results of our experiments that lead us to claim that accurate algorithms may be applicable with realistic performances in the X-domain context.

End-to-end quality of service in pseudo-wire networks

Carrier-grade networks are complex systems that include several heterogeneous domains and support various types of services under specific Quality of Service (QoS) requirements. To tackle the problem of setting end-to-end connections across heterogeneous domains, the Pseudo-Wire architecture [1] allows to emulate some protocols (e.g. SDH, Ethernet, ATM, etc.) over MPLS. This emulation is achieved by encapsulation and decapsulation functions called adaptation functions. A path crossing heterogeneous domains must involve compatible functions so that datagrams are understandable by the source and target nodes (e.g. if Ethernet is encapsulated in MPLS by a node, it must be decapsulated by another).

Distributed E2E QoS-Based Path Computation Algorithm over Multiple Inter-domain Routes

Internet usages have changed with the emergence of value added services relying on a higher interactivity and needs for a better quality of experience (QoE). Telecommunication operators have to face a continuing growth of new types of Internet traffic (video, games, telepresence, etc.) imposing not only a more efficient utilization of their network infrastructure resources, but also the generation of new revenues to pursue investments and sustain the increasing demand. Such services generally cross multiple domains, but inter-domain routing protocols still have some limitations in terms of service assurance. For example, BGPs single route announce for a destination limits potential traffic engineering features (e.g. no quality of service price/efficiency optimisation, inter-domain shared route protection, inter-domain load balancing, etc.). In order to provision end-to-end inter-domain connections that obey to constraints such as bandwidth, delay, jitter, and packet loss for these services, an interesting approach is to compute end-to-end (e2e) paths over multiple inter-domain routes. This will allow establishing more efficiently the inter-domain connections with respect to requested QoS constraints and sharing these constraints (and associated revenues) among multiple operators to globally accept more demands in the system, while keep satisfying the customer QoE. To address these challenges, we propose an efficient distributed inter-domain algorithm that computes such constrained paths among a set of domains, exploring multiple inter-domain routes. We demonstrate that our algorithm not only increases success rate in delivering feasible paths, but also admits more connections and keeps a reasonable runtime.

Testbed implementation of control plane extensions for inter-carrier GMPLS LSP provisioning

This paper presents a testbed implementation of an inter-carrier GMPLS (Generalized Multi Protocol Label Switching) service architecture recently proposed. This architecture couples the Path Computation Element (PCE)-based control plane with a service plane managing discovery, composition and activation functions of inter-carrier service elements. The testbed implements the required PCE Communication Protocol (PCEP) and Resource Reservation Protocol with Traffic Engineering (RSVP-TE) extensions, together with service request filtering operations performed with a policy based architecture1.

A protocol for QoS contract negotiation and its implementation using Web Services

The way Internet is used changes: demand grows for critical services that cross several provider networks; guaranteeing a required end-to-end quality of service (QoS) across several networks becomes a challenge. Some critical services (e.g. video-conference, VPN etc.) can not be satisfied in a best effort fashion. The use of QoS contracts (service level agreements, SLAs) is effective for management of such services. However, the problem of meeting end-to-end QoS requirement remains: no centralized entity can compute overall QoS for chains of contracts, and a fortiori, such contract chains can not be optimized centrally. Thus, the following problem has to be solved: given an end- to-end QoS request and collections of available SLAs on each participating domain, establish an end-to-end contract committing a chain of providers and giving optimal service under most reliable guarantees available.

Path computation in multi-layer multi-domain networks: A language theoretic approach

Multi-layer networks are networks in which several protocols may coexist at different layers. The Pseudo-Wire architecture provides encapsulation and decapsulation functions of protocols over Packet-Switched Networks. In a multi-domain context, computing a path to support end-to-end services requires the consideration of encapsulation and decapsulation capabilities. It appears that graph models are not expressive enough to tackle this problem. In this paper, we propose a new model of heterogeneous networks using Automata Theory. A network is modeled as a Push-Down Automaton (PDA) which is able to capture the encapsulation and decapsulation capabilities, the PDA stack corresponding to the stack of encapsulated protocols. We provide polynomial algorithms that compute the shortest path either in hops or in the number of encapsulations and decapsulations along the inter-domain path, the later reducing manual configurations and possible loops in the path.

The price of tussles: bankrupt in cyberspace?

The question of who funds the Internet network infrastructure is a timely and hot debate. The debate recently involved a tussle between two values: the content against its carriers, i.e. any operator providing the data transport service, let it be a transit or an access Internet Service Provider (ISP). On the one hand, carriers, alarmed by the data traffic explosion [1], and seeing their sources of revenues diminish (see Sec. 2.2), complain that most of the Internet revenues are unfairly seized by few content and application providers. They want the latter to share the costs of the network infrastructure. On the other hand, content providers can rightly argue that, as users’ prime motivation for buying the Internet access is to retrieve and share content, they should have in return a part of the Internet access revenues. Unfortunately, such tussles exceeded the stage of mutual accusations and press statements to emerge in the form of several “peering disputes”, and practices like ISPs blocking certain applications, with sometimes consequences on the users’ quality of experience. Clark et al. [2] foresee such tussles as those that will draw tomorrow’s Internet. But will there be a tomorrow for the Internet without a network infrastructure? 2016 is seen by many as a breaking point in the Internet economy. Under the assumption that current trends in prices and demand will hold, revenues will not be able anymore to support the required investments in network infrastructure. But who foretells the future lies even if he tells the truth; regardless of whether 2016 will be a “black-out” or not, it is important to accommodate such tussles [2] instead of ignoring them. Reconciling the content with its carriers needs an understanding of the deep reasons that lie behind this conflict. We argue that this network infrastructure funding tussle hides a deeper problem: the obsolescence of the Internet early-days business models and their inadequacy to adapt to the changes that the Internet underwent. We depict two of these changes and show how the legacy business models failed to accommodate them. We conclude by discussing possible approaches to de-ossify business models and stress their limitations.