Interaction of overlay networks

Wenjie Jiang, John C. S. Lui, Dah-Ming Chiu
2005 Performance Evaluation Review  
Although the concept of application layer overlay routing has received much attention lately, there has been little focus on the "co-existence" and "interaction" of overlays on top of the same physical network. In this paper, we present some fundamental insights of these overlay interactions. First, we show that when each overlay performs the overlay optimal routing so as to optimize its own performance, there exists an equilibrium point for the overall routing strategy (in other words,
more » ... al routing optimization will not lead to a routing instability). However, the equilibrium point may be inefficient. We discover some implications due to the interactions: (a) the equilibrium point may not be Pareto optimal, (b) some fairness anomalies of resource allocation may occur. We further show the "performance gap" to the global optimal routing policy, which is often used by an ISP to control the underlying traffic. This is worthy of attention since overlay may not be aware of the existence of other overlays and they will continue to operate at this sub-optimal point. Thirdly, we explore two distributed pricing schemes to resolve the above issues. We show that by incorporating a proper pricing scheme, we can lead the selfish behavior of overlays to an efficient equilibrium. Extensive fluid-based as well as packet-based simulations are carried out to support our theoretical claims. We believe this work provides fundamental understanding of overlay interactions, and stimulates further research on this issue. find a better route to ensure QoS guarantee service. From the overlay's point of view, application routing is a form of optimization in which the overlay tries to maximize its desired objective, such as reducing the delay and/or cost, subjected to the availability of the network resources. Overlay networks is not a new concept. The Internet can be viewed as an overlay network, built largely on top of telephone and other (e.g. frame-relay, ATM and others) networks. The Internet's underlay provides specific bandwidths, therefore emulates physical wires. When building overlays on top of the Internet, however, the logical links between nodes are based on Internet's best-effort service. Consequently, the performance of Internet overlays will depend on how it co-exists with (a) the existing Internet traffic, and (b) the traffic of other Internet overlays sharing same physical links. Although the concept and deployment of application layer overlay networks have received much attention recently, there has been little focus on (a) the "co-existence" and "interaction" of multiple overlays, (b) the "control" of this interaction so as to ensure the overall network performance and/or fairness. In here, we are interested in the scenarios in which multiple overlays are constructed above the same physical network. These overlays have partially overlapping paths and even nodes. Since overlays do not know the existence of other overlays, and meanwhile, each overlay tries to optimize its own performance without considering the impact on other overlays. Therefore, there is an inevitable interaction between these overlays due to their respective optimization actions. This form of interaction has some intrinsic properties as well as some undesirable effects (i.e., degradation of overall network performance, fairness anomaly in resource allocation, which we will present in Section 5). To control these effects, pricing scheme is proposed to reduce or eliminate these effects. Note that the pricing scheme can be used by the ISPs, not only to increase their revenue, but also as a mean to perform traffic engineering so as to achieve the global optimality, or improve fairness. The contributions of our work are: • We present and formalize the "overlay optimal routing" policy under an optimization framework. We illustrate its relationship with the "global optimal routing", and show that there exists a performance gap between two routing policies. • We model the interaction of overlays as a non-cooperative game and show that there exists a Nash equilibrium (i.e., implying that there will be no routing instability) under a general network setting. • We report a number of anomalies due to the interactions between overlays. Firstly, the equilibrium point is not Pareto (or social) optimal, which can cause a "tragedy of the commons", meaning that the performance of all overlays can be seriously degraded. Secondly, an interesting and more important discovery is that the interaction may lead to some fairness anomalies in resource allocation. That is, at the equilibrium point, it is possible for some overlays to obtain a higher percentage of the common resource (e.g., link bandwidth) as compared to other overlays, and cause these overlays to experience a significant performance degradation. • To overcome the above two issues, we propose two pricing mechanisms to alleviate the problem. We show that by incorporating a proper pricing scheme, we can either improve the overall performance, or bring in the fairness in resource allocation, and at the same time, increase the ISP's revenue. • We illustrate via fluid-based and packet-based simulations to support our theoretical findings. The balance of the paper is as follows. In Section 2, we present the overlay optimal routing policy and its mathematical optimization model. We study the interaction between co-existing overlays as a non-cooperative Nash routing
doi:10.1145/1101892.1101902 fatcat:rusehs5m3bczxkblx7rl5mkxsm