CAPITULO IV RESULTADOS Y DISCUSION
4.2. Prueba de Hipótesis General
4.2.1 Prueba de Hipótesis especifica
The resource management domain encompasses a broad range of research. Since the emergence of multihoming in mobile devices, areas pertaining to the effective use of the additional network resource have become more prominent. In this section we will review the various approaches to the management of network resource in multihomed mobile devices.
3.3.1
Always Best Connected
The Always Best Connected (ABC) model for connectivity was one of the earliest concepts for management of multihoming in a mobile context. In [68], the authors present the concept of the ABC architecture and the associated challenges from both a user and business perspective. The base premise of ABC is based on exploiting diversity in access technologies and access providers to ensure that the
user is always connected to the optimal network.
The authors of [68] present a set of components that are required to sup- port ABC, including: access discovery, access selection, AAA support, mobility management, profile handling, and content adaptation. While the definition of the components remains abstract the proposal and design is well grounded in addressing the challenges of the mobile domain and is of significant value. Access discovery requires a host to discover the available access networks, periodically updating this information to ensure there are no better alternatives. During the discovery stage, the authors propose that the available access networks pro- vide the host with a set of parameters describing the quality and nature of the access network. Access selection presents both host based and network based approaches. The host based selection model, would allow the host to draw from user preferences to make a selection decision. A network based approach would take a broader, possibly centralised view; with a network service specifying the best access network for a host to use, which could help to alleviate congestion, by balancing users across different access networks. AAA support, is required to ensure both users and providers are protected, this is of significant importance if connecting via another user or roaming on an unknown network. Mobility man- agement becomes even more important for mobile devices in the presence of ABC, as users may regularly switch between points of attachment as the state of net- work access changes, without appropriate mobility protocols, this would disrupt the users connections and degrade the quality of service; appropriate mobility models have been discussed in Appendix A. Profile management is presented in the context of an ABC provider hosting a users preferences, such that access net- works are always aware of the users needs and requirements. Finally, the content adaptation component specifies that applications should be able to adapt to the current conditions. The adaptation may take form by providing applications with hooks into the ABC model; for example, a video may change the desired bitrate based on the current state of connectivity.
The work presented by Song et al. in [159] proposes a scheme to ensure users receive the best possible QoS at any given time. To this end, their solution is comprised of two components, they first evaluate the criteria and weightings of the QoS metrics, secondly the authors use grey relational analysis to rank the choices. This process attempts to balance the quality of the network against user preferences while also limiting frequent handoffs. Determining the most appropri-
ate network at any given time is a non trivial problem. As previously mentioned, the authors propose balancing network metrics with user preferences, which re- sults in a multi-parameter or multi-objective optimisation problem. Therefore the use of well established mathematical models in such a scenario can help to simplify the problem domain.
Wilson et al. presented an alternative approach to optimise wireless access in [172]. Due to the complexity and uncertainty associated with choosing the best network, the authors propose the use of fuzzy logic to infer the optimal access network. This is justified in part, due to the subjective nature of po- tential QoS parameters; for example, users may be able to specify criteria such as “low latency” or “low cost”, which makes the definition of an appropriate multi-parameter model very difficult. The use of fuzzy logic allows the authors to account for this ambiguity in requirements. As discussed by the authors, the core problem with this approach for network selection is that it requires a simple rule base, that must be built in advance. The addition of alternative QoS parameters may become challenging and overly complex, to counter these limitations the au- thors propose the use of learning algorithms, with the ultimate goal of matching the users behaviour.
3.3.2
Policy Based Network Management
In the realm of ABC, the authors typically discuss the use of “user profiles” or prior configuration on the end-host. Leveraging user policy to describe the ap- proach to connectivity can help to minimise overly complex selection algorithms. A current example of such a tool is the Tasker [51] automation application for Android. Tasker simplifies the ABC model, by requiring the user to specify their network connectivity options based on a set of events and actions. For example, Tasker may react to battery capacity dropping and subsequently turn off the WiFi or Cellular interfaces. While Tasker is not built as a dedicated network management application, the ability to turn network interfaces on and off based on pre-defined events improves the basic connectivity model for mobile devices. In addition to simple applications, policy based network management has been a significant research problem in the core of networking, relating to service layer agreements, traffic management and ensuring QoS. Typically the policy based solutions focus on the core of the Internet and are of limited use at the edge,
leaving the end-host to make decisions for itself.
The user profiles specified in the ABC domain are typically simple; for exam- ple, prioritising access to one network over another. In [116], the authors propose putting the user and their associated context at the centre of the selection prob- lem. To this end, the authors define a set of rules that can be used to limit energy consumption, for example, if their are no on-going communications turn off non-cellular interfaces, unless a user manually intervenes. By putting the users requirements at the core of the system, the multi-parameter optimisation prob- lem can be addressed by a simple additive weighting model [1]. Additionally, the contexts the authors propose can increase the granularity of policy definition and improve network selection beyond looking at QoS. This is especially beneficial for applications that can adapt to the available bandwidth or network quality.
3.3.3
Path selection
The work presented in the ABC domain, has typically focused on the use of a single interface at a time. In the previous chapter, we introduced a range of resource pooling protocols that break this model, and rely on the simultaneous use of multiple network interfaces to maximise performance. To this end, path selection algorithms may provide a more fine grained approach to ABC. The general scope of the path selection problem is similar to ABC, as there is still a multi-parameter problem to solve, to define the best path.
In [93], the authors propose using active network measurements to help inform the path selection decision, sending probe traffic to gauge the RTT and the bot- tleneck bandwidth. This work focuses on addressing multihoming issues within SCTP, therefore each connection has a primary path and a set of backup paths. Given the set of paths and the associated metrics gathered by the probe traffic, the authors propose a simple set of rules to determine the most appropriate path. The rules comprise of a set of “if-statements”, looking to determine which net- work path has the higher bandwidth, if both paths have sufficient bandwidth the round trip times are compared. As the solution is bandwidth driven, there is a possibility for a high rate of churn, as multiple SCTP connections may attempt to converge on the same path.
In [61], the focus is again on the multihoming properties associated with SCTP. The authors propose Wireless SCTP Extension (WiSE) to improve the
efficiency of an SCTP connection, by ensuring the best possible path is always used. Three extensions to SCTP are presented, focusing on: congestion control, path management, and bandwidth estimation. Firstly the authors attempt to distinguish between losses due to congestion on the path and losses due to bad channel conditions. This is achieved by comparing the current output rate of the connection to the last known estimation of bandwidth. If the output rate is higher, the authors assume the loss is due to congestion, while if it is lower the physical channel is held responsible. The path management approach is very simple, purely focusing on the bandwidth of each path. If an alternative path has a higher bandwidth than the current path, a switch will occur, resulting in the alternative path becoming the new primary path. Finally to estimate bandwidth the authors use both active and passive measurements, adding additional com- plexity in comparison to [93]. The primary path uses passive measurements due to the assumption that there will be guaranteed traffic, this simply compares the amount of data confirmed by the SACK and the RTT, to estimate bandwidth. As no traffic is sent on the backup paths, active network measurements are used, relying on sending a train of packet pairs as proposed in [76].
The work presented in [91] takes a QoS based approach to network selection closer to that presented in the ABC domain. The authors simplify the QoS opti- misation problem by selecting a minimal set of parameters, including: monetary cost, area of coverage, required bandwidth, and the number of network interfaces to be connected. Furthermore, it is proposed that users be split into three cate- gories: bronze, silver, and gold. Bronze users prioritize cost over QoS. Silver users desire a fair trade off between QoS and cost. While gold users demand the best QoS regardless of the cost. The authors then propose using an analytic hierarchy process to provide weightings for each QoS category for each user. Subsequently each path is provided an overall utility value based on the weight of criteria. The Euclidean distance is then used to determine which path is closest to the optimal values. The presented QoS metrics may not be sufficient in all cases to describe a users needs. As the authors focus on bandwidth and cost, important quality metrics such as loss, and delay are ignored. In the case of voice applications and other real time applications, high bandwidth is much less important than stability and delay. While there may be a clear relationship between loss, delay, and bandwidth; this can easily be broken for a number of reasons, such as access points and providers placing bandwidth limits on individual hosts.