selection of QEAPs, SPAD enables the provision of QoS on the connections between the elementary application components that form an overlay application.
An extensive performance evaluation of SPAD was provided in chapter7. This evaluation was based on simulation experiments, using measured data from three different community of peers on the Internet. These experiments indicated that: i) the SPAD QEAP discovery schemes succeeded at discovering a significant amount of existing QEAPs within each considered community, and ii) the SPAD QEAP selection scheme succeeded at deciding which QEAP best met the user’s QoS preferences within a set of discovered ones.
Thus, this thesis demonstrated that it was feasible to provide QoS to an overlay appli- cation by using alternate Internet paths resulting from the compositions of independent consecutive paths. Furthermore, it also demonstrated that it was possible to discover and compose these independent paths in a distributed manner within an community comprising a limited large number of autonomous cooperating peers.
8.3 FUTURE RESEARCH DIRECTIONS AND PERSPECTIVES The proposed SPAD system has been designed and evaluated for a bounded community of peers, such as the community of service providers as introduced in chapter2. Compared to the entire Internet, this community has a smaller number of hosts. The generalized use of QEAPs and the deployment of distributed QEAP discovery/selection schemes on all hosts of a vast network like the Internet is not within the focus of this thesis and requires further studies. Such studies possibly involve
• an extensive evaluation of the SPAD schemes in a dynamic environment (as suggested and discussed in section7.6),
• the design of an advanced admission control scheme to be deployed on potential relay nodes (as suggested and discussed in section6.5),
• the design of an adequate congestion control scheme to ensure that SPAD traffic coexists in a friendly manner with other type of traffic, e.g. TCP traffic. Such a congestion control scheme can operate independently on each hop of an overlay as- sociation, or as a global scheme on an entire overlay association. Moreover, the same choice also occurs within an given QEAP (i.e. a hop of an overlay association), where congestion control can also be deployed independently on each path components, or globally on the entire set of paths composing the considered QEAP. The selection between these types of possible schemes requires further investigations.
Lua et al. [107] propose a taxonomy of peer-to-peer systems with two main categories, namely systems based on a structured or an unstructured model. As stated in section4.4.3, the proposed SPAD system is based on the unstructured model. This design decision is arbitrary and further studies are required to decide which model is more adequate to QEAP discovery and selection. In the early stage of the research work related to this thesis, a very basic QEAP discovery scheme based on the structured model was proposed [108]. The scheme presented in section 6.2inspired this basic structured scheme, which is also based
on a community of cooperating peers. It proposes to embed these peers within a virtual space according to their relative delays, and to use a modified distance function on a given peer NA to compute/discover potential QEAPs [NA → NX → NB]. The performance of
this scheme is significantly limited by the non-metric nature of the latency QoS parameter2. Some existing contributions provide methods to embed non-metric pairwise proximity data into virtual metric spaces, without losing any existing clustering properties among the considered data sets. The “constant shift embedding” technique from Roth et al. [135] is an example of such a method. Investigating the feasibility/performance of a structured QEAP discovery scheme, which possibly uses such an embedding method, constitutes a challenging potential research direction.
This thesis proposes the use of single virtual Internet paths to enhance the Quality of Service between pairs of Internet hosts. The study of other type of applications for these virtual Internet paths is another interesting research avenue. For example, one can use mul- tiple virtual Internet paths coupled with network coding and/or Forward Error Correction (FEC) methods to maximize the utilization of the overlay links between nodes involved in simultaneous overlay applications. In another example, one can use multiple virtual paths in conjunction with multicast schemes on overlay networks to enhance the QoS perceived by the receiving peers. In these examples, the use of multiple virtual paths also raises a fairness issue. When a user A sends its data traffic over multiple virtual paths, he/she utilizes more network resources than a user B who only require one path. This result in a potential fairness issue, when for example a congestion event occurs on a given network point (i.e. router) shared by these users, the competition between the traffic from user A and B might potentially be biased towards user A. Ensuring the fairness of these multiple path schemes with the currently existing communication schemes is another interesting challenge.
Furthermore, this thesis also proposes the use of virtual Internet paths that enhance only two particular QoS parameters, namely latency and packet-loss rate. However, there exists many applications that require QEAPs, which enhance other types of QoS parameters (such as jitter, bandwidth), or a complex constrained combination of multiple QoS parameters. Several works [82, 83, 84] studied this multi-constraint path problem in the context of QoS routing (as presented and discussed in section 4.3.1), but few contributions studied such a problem in the context of overlay networks and overlay paths. As introduced in section 3.3, the overlay network model virtualizes an existing network infrastructure for the above users. Such a feature can potentially lead to the design of new multi-constraint path selection schemes, or allow the deployment of modified existing QoS routing schemes. The design, modification or deployment of such multi-constraint routing schemes in the context of overlay network constitutes another possible research direction.
Finally, one can also study the feasibility of using the SPAD system as the base of a business-oriented service, similarly to QRON [60] and OverQoS [48]. For example, an association of business partners might form a SPAD community, and act as a gateway to provide or sell QEAPs to other Internet users. This business-oriented context raises other potential research problems such as the type of Service Level Agreement (SLA) to use
2As demonstrated in chapter5, Internet delays do not necessarily satisfy the triangle inequality, hence