Context-aware framework for adaptive routing

Introduction

The intensive use of the current IP networks lead to the proposal of different solutions in order to enable multiple types of traffic to coexist and optimize the utilization of resources as much as possible within the existing network infrastructure. Meanwhile, emerging applications and services generate an increasing amount of real-time data traffic to the network. Unfortunately, network infrastructure and routing strategies have not evolved at the same pace as data applications. Thus, network infrastructure is constantly under resources shortage and consequently under congestion.

Along the last 20 years, different contributions have been proposed in the literature in order to provide real-time traffic guarantees and fairness concerning the utilization of network resources. Therefore, providing Quality of Service (QoS) has been the key solution in order to meet user´s expectation. For this purpose, contributions such as Integrated Services (IntServ), Differentiated Services (DiffServ), MPLS/GMPLS and Traffic Engineering have been largely deployed [1].

In particular, service providers have deployed QoS techniques when determining configuration strategies, planning and provisioning network services. These techniques were related, in general, to admission and congestion control, buffer management and scheduling. Nevertheless, QoS has been proven insufficient in order to meet user´s expectations considering the delivery of real-time sensitive traffic. In many situations, regardless of the techniques applied, the expected quality is not achieved completely, generating consumer dissatisfaction with the services offered [2]. Indeed, Service Level Agreement (SLA) establishes users and infrastructure parameters for the delivery of a particular traffic, such as the QoS metrics related to the dynamic nature of the network (such as bandwidth, delay, jitter, among others). However, the dynamic nature of the user and application´s environment should also be considered.

In order to embody the concept of a user-centered network the notion of context can be applied. Context awareness is understood as an ubiquous and/or pervasive computing paradigm that aims at dealing with changes in the computational system environment [3]. The implementation of context aware networks can be helpful in order to improve user´s experience and satisfaction when accessing network resources. For this purpose, this work addresses the proposal of a context-aware framework, which relays on three main concepts: Quality of Service, Quality of Experience and Adaptive Routing. This paper is structured as follows: Section 2 discusses some contributions in the literature concerning the proposal of QoS-based frameworks, adaptive routing and user experience; Section 3 introduces the proposed context-aware framework; Section 4 presents some conclusions and future perspectives.

Related Works

Different issues should be considered when addressing context-based routing. Context is related, in particular, to both user and communication platform. Therefore, the main aspects discussed in this paper in order to address context are frameworks proposed to manage Quality of Service, adaptive routing and definition and implementation of Quality of Experience.

QoS Management Framework

Quality of Service (QoS) has been conceived as a solution in order to provide different traffic guarantees to different applications, users or data flows. Some of the metrics applied to implement QoS are bit rate, delay, jitter, among others.

Several QoS paradigms have been proposed, such as Integrated Services which provides guarantees for individual flows, Differentiated Services which implements classes of services in routers for aggregated traffic, MPLS which provides specific routes for the delivery of flows and better utilization of bandwidth resources, GMPLS which is an extension of MPLS to support optical networks [2], etc.

Based on the experience acquired with these contributions, different frameworks for managing QoS have been proposed in order to solve specific issues both proposing a decoupling between network control plane from the data plane and presenting an underlying model developed as a QoS framework. The design of some of the existing QoS frameworks in the literature followed a pattern related to the definition of four main modules:

• Management of network/application resourceswhich is responsible for negotiating Service Level Agreement (SLA) with customers and communicating SLA related parameters to the respective resourcers manager, besides guaranteeing SLA according to the assigned resources [4] In most of these contributions the authors applied the same set of metrics in order to propose their respective QoS frameworks, such as latency, throughput, failure and cost. Other more specific metrics are also applied such as availability, security, accessibility and regularity.

Besides the existing efforts in order to provide resources availability and priority, it is also important to provide means of optimizing the existing routing paths according to the state of the network. Therefore, we can also consider different contributions in the literature in order to provide adaptive routing.

Adaptive Routing

In recent years scientific contributions have also focused on the proposal of different solutions for the dynamic and adaptive routing of time sensitive data traffic. One of these solutions is the deployment of traffic engineering, which has been widely applied in order to provide a balance between bandwidth availability and the increasing demand of traffic. Another important related issue in order to enable real-time traffic deployment is congestion control, which allows the optimization of buffers occupation.

These approaches have been applied as solutions for adaptive routing with some minor variations depending on the proposed model or algorithm. In general, the existing contributions can be classified as follows:

 Based on context [11] [12] [13];  Based on traffic engineering [14] [15] [16], and;  Based on probabilistic data [17].

In general, the existing context-based adaptive routing approaches apply distinct metrics compared to those applied in Quality of Service approaches, such as delay, jitter, etc. In opposite, the algorithms and solutions implemented as context-based protocols, in general, have deployed some other metrics such as connectivity change in a host, power level, delivery probability, signal strength, health node, hop counter and learning rate.

The CAR algorithm [11] has been proposed based on context information in order to avoid packet flooding and bursty traffic. This algorithm proposes a proactive routing based on the highest delivery probability. Thus, the routing path can be adapted if there is a host with higher delivery probability. This algorithm has been implemented for wireless ad-hoc networks. Another context-based routing contribution introduces a framework that aims at studying the behaviour of wireless sensor networks [12], however it can also be applied to any other type of network. In this work, the metrics applied in the context information are: node selection criterion, constraint routing rules, etc. At last, the work presented in [13] introduces a context-based framework for delay-tolerant networks. In this contribution, an adaptive gateway is described based on external context obtained by agents, which is able to modify the internal network routing.

The solution proposed in [14] introduces TeXcpas an on-line distributed traffic engineering protocol. This protocol reacts to real-time events in order to provide the optimization of the deployment of a low bandwidth link and meet the growing demand of traffic. The work presented in [15] introduces an algorithm for dynamic distributed traffic engineering called Replex. This algorithm is based on a re-routing policy in order to avoid oscillation and enable rapid adjustment in case of convergence. In a more recent study [16], the authors also address the adaptive routing based on the offline provisioning of multiple paths and the online load balance during routing planning.

The solution called Sight [17] proposes a routing scheme based on probabilistic information distributed for self-adaptive routing in real dynamic environments. Routing with Sight is based on the balance of the link, latency and link utilization in order to avoid possible overloads. In this solution, some parameters are used such as: latency, delay, contention at the MAC layer, utilization of the link, among others.

When studying and proposing solutions for the optimization of routing it is important to consider Quality of Service (QoS) and Quality of Experience (QoE). As expected, the contributions related to QoS frameworks focus on the proposal of protocols and mechanisms in order optimize the resources availability related to network equipments. QoE, instead, refers to users´expectations and how they actually perceived the service delivered. In order to meet user expectations, the implementation of QoS should be centered on the perspectives of the end users, to ensure that quality of service meets the levels of expected QoE.

Quality of Experience

Although QoS has been largely deployed, QoS metrics have been proven insufficient in order to meet user´s expectations considering the delivery of real-time sensitive traffic. Assessing user´s satisfaction from customer´s perspective is somewhat a complex task since it requires the definition of metrics that are not only related directly to the classical QoS parameters. Indeed, well-defined QoS policies and rules are not an absolute guarantee of user´squality experience.

Quality of Experience (QoE) is the evaluation of subjective perception from user´s perspective. In opposite to QoS, which is network-centric, QoE is user-centric and is based on cognitive, behavioral and psychological factors [18]. Therefore, it is related to the study of the perspective and understanding of user perception, contemplating expectations and the experience upon a particular application.

The challenge related to understanding QoE is to determine which metrics can be applied in order to provide measuring, mapping, anticipating and predicting user´s subjective perception. For this purpose, QoE can also be understood as describing four areas: contextual, technological, business and human [19], remaining as a pseudo-layer between the application and the network [20].

QoE Metrics

In general, QoE can be correlated by the measurement of MOS (Mean Opinion Score), whose values range among bad experience, poor, acceptable, good and excellent [2]. This is a subjective measure that can be obtained by means of a satisfaction survey or by inference gathered from the correlation with other parameters. When it comes to multimedia the Peak Signal Noise Ratio (PSNR) is a parameter often adopted to qualify QoE, and their corresponding MOS [21] [22]. Another approach is based on the Perceptual Evaluation of Speech Quality (PESQ), which is a family of standards based on voice test samples defined in the PESQ application guide ITU-T P.862.3. and its mapping/correlation with MOS [23].

The existing contributions in the literature aim at mapping QoS parameters and other existing parameters related to the analysis and assessment of video quality into the corresponding QoE by means of mathematical formulas and already existing parameters for the evaluation of video quality as proposed in [23]. Some contributions consider that the nature of QoE perception is based on the correlation between QoS and QoE, involving existing parameters, which are already applied to the measurement of multimedia applications.

Besides the existing contributions that aim at measuring objective parameters and further mapping them into QoE metrics, other contributions also collect information directly from users using techniques such as crowdsourcing [24]. Another approach also makes usage of "probes" strategically distributed [2]. Although there is no dominant single approach established, some strategies adopted metrics obtained from video quality parameters, enabling effective measurements experiments in order to correlate QoE and MOS. The correlations that are mostly applied are PESQ/MOS and MOS/QOE [23]. [20], social context, concentration [18], and background noise [18] [27], among many other human factors.

This section outlined some of the existing contributions in the literature in order to correlate Quality of Experience with the familiar MOS (Mean Opinion Score) and try to correlate the parameters of MOS with video quality parameters (PSNR, PESQ). Besides the QoE metrics identified in this study, it is also important to consider contextual data, which can be applied for improving the delivery of services and thus maximizing user experience.

Context Awareness

Context can be defined as "any information that can be used to characterize the situation of entities (i.e. whether a person, place or object) that are considered relevant to the interaction between a user and an application, including the user and the application themselves. Context is typically related the location, identity and state of people, groups and computational and physical objects" [28].

Figure 1 -Five Fundamental Categories for Context Information [29]

The use of context can be applied to entities, persons, places, or even to an object relevant to the application, by the definition of characteristics of individuality, activity that may be involved, location and time and even relationships with other entities [29], as depicted in Figure 1.

The definition of context is also related to the description of Contextual Elements (CEs). While context refers to the interaction of an agent and an application, CEs characterize the domain of the context that this agent is inserted [30].

Thus, context-aware applications are able to provide services with assistance-based tasks, contextbased actions and adaptation of behavior of the system according to contextual information [30] [31]. These services are so-called context-aware services (CASs) [32].

In order to assess the information described by the context, and the device which will be the source of this information, the notion of Quality of Context and Quality of Device are considered.

Quality of Context

Although the contribution of context-aware systems can be expressive, their effectiveness can only be achieved if context information is properly defined. Therefore, the definition of Quality of Context (QoC) is required in order to provide the understanding between QoC, QoS and Quality of the Device (QoD). The latter is related to the hardware components involved in providing the context information [33] [31].

In order to provide QoC some metrics should also be defined [33]: Accuracy of information (Precision); Likelihood the information is correct (Probability of Correctness); Level of trust in sources of information (Trust-worthiness); Resolution of the levels of granularity of information (Resolution); Timeliness of information related to their temporal characteristics (Up-to-dateness).

Nevetherless, it is important the adoption of clear policies in order to provide the correct analysis of contextual information and to be in conformance with QoC [34]. For this purpose, some contributions in the literature proposed solutions for improving the adoption of QoC policies based not only on the current context, but also on the effects of erroneous context information with low quality and its effects on systems, such as Proteus [35].

Quality of Device

Besides Quality of Context (QoC) concerning the characterization of the collected context information, it is also important to consider the Quality of Device (QOD), which is related to the precision of the computing device that will collect the context information. For instance, the Global Positioning System (GPS) of each device can have different levels of precision, or even a particular device which is not to provide some parameters compared to another due hardware incompatibilities or the lack of ability to collect such information [33]. Therefore, QoD will provide information on the technical characteristics of each device and its capabilities [31].

Context-awareness: Related Works

Context information allows systems to become more assertive, thus requiring less effort from users to provide information. Nevetheless, context information can also be useful for providing support to decision making when it comes to routing data within the network. In this case, the context can be defined through periodical measurements carried out by the nodes of a network in order to assess connectivity, resources availability (bandwidth, queueing, etc.), etc. in order to meet QoS specifications and also proposing alternative routes, as in the case of opportunistic networks [12] [13]. Therefore, decisions can be made based on user's context, its connection partners, their computational characteristics, among others.

A comparative study of some contributions related to context-aware routing was carried out [13] [36] [12] [37] [11]. Most of these contributions are related to the utilization of context applied mainly to wireless networks. In these studies, the use of context allowed improvements mainly in: stability of the communication link, increased bandwidth (by decreasing overhead), greater autonomy of batteries, shorter delay and gain of scale. These environments differ greatly from wired networks, mainly due to storage capacity and processing constraints, battery life limitation, and in some cases, limited bandwidth.

Proposed Solution

The definition of context information and context-aware routing enabled the proposal of a generic context-aware data routing mechanism. Thus, network devices such as routers and switches are able to choose properly the routing paths and traffic prioritization based on context information. This section presents an overview of the proposed context model and the struture of the proposed framework.

Context Model

The proposed context model that will be applied in the context-aware framework for adaptive routing can be used for both wireless and wired networks. This model is generic allowing the description of different network scenarios and the adaptation based on user´s experience. The adopted context model describes the state of a particular entity (for instance, a user, router, switch, etc). In general, generic features can describe this entity, such as ( Figure 2):

 Individualitywhich describes a particular information about an entity, such as identification, addressing, protocols, etc.;

 Timewhich describes time information, such as timestamp in order to allow registering the status of an entity in a given time;

 Locationwhich is related to real or virtual location of an entity, and may be generated by a system such as GPS location, or by referencing information such as home, building, city, a network address, etc;

 Activitywhich allows the description of explicit goals, tasks and actions performed by an entity, and;

 Relationswhich describes the entity's relationships with other entities, dependencies between entities, connections with objects, people, places, services, etc..

Moreover, some other aspects can still enrich the description of an entity such as:

 Quality of Experience (QoE)which describes a group of parameters regarding user´s perspective, that most of times could be rated as MOS.;

 Quality of Device (QoD)which describes a group of parameters regarding devices characteristics like capabilities, computational power, precision level of data colectors, and;

 Quality of Service (QoS)which is related to all the metrics (qualitative/quantitative) considered on an SLA between user and platform, such as (bandwidth, delay, jitter, etc.).

Figure 2 -Context Model

The context model can also be validated according to some metrics, which determine the Quality of Context:

 Precisionlevel of information accuracy to assess its relevance;

 Probability of correctnessassessment of the probability of the information being correct;

 Trustworthinessassessment of the level of trust on the information source;

 Resolutionlevel of granularity of a given information;

 Up-to-dateness -assessing how the information provided is updated.;

The following excerpts illustrate the XML-based description of an entity, respectively for a device, a user and an infrastructure perspective (QoS).

 XML for a Device: The modular proposal for this framework allows components to be developed independently, even though they apply and manage the same common representation for the context information, as defined in XML description presented previously.

Context-Aware Routing Framework

The solution proposed for context-based routing has been designed based on the integration of different functional modules. Thus, the Context-aware Adaptive Routing Framework (CARF) has been proposed in order to meet the following requirements:

 Collect and share context information from/among key devices connected to the network (e.g., nodes);  Centralize context information storage;  Ensure QoC policy analysis before the release of context information, and;  Query context information interface, with flexible input and output definition, enabling context-aware services based on network to use such information.

In order to meet these requirements, the following modules were proposed: o It supports the dynamic reconfiguration of flow tables into the forwarding devices (switches).

The structure of the proposed framework is illustrated in Figure 3.

Figure 3 -Structure of the proposed Framework

It is important to note that Context Routing Manager relies only upon the context information in order to create flow tables. Therefore, the proposed framework is able to provide alternative routing techniques that can be applied to any type of network in order to reach satisfactory levels of quality of service and quality of experience.