ITR: Collaborative Research: Resource Allocation and Denial of Service Prevention in Active Networks

点击翻译按钮获取中文摘要

The Internet is used by a rapidly expanding and changing set of applications. The need for the network to evolve and even to provide application specific processing is significant. However the current network infrastructure is hard to evolve and does not readily support customizability. The goal of Active Networking [21, 3, 2] is to facilitate this evolution and customization by making the network infrastructure programmable. One way of adding programability is to allow code to be down-loaded into the routers, thus enabling the addition or modification of services. A more radical approach is to allow the packets themselves to carry programs to be executed selectively on the network's routers. Among other issues, these two approaches increase the possibility of denial of service attacks whereby a user places excessive demands on network resources in order to deny access to another user. However, they also enable new approaches to handling such attacks and to addressing the general problem of allocating resources within the network.The proposed research focuses on issues involving programmable, or active, packets. Active packets facilitate denial of service attacks in several ways. First, unlike conventional data transport packets, an active packet may require processor cycles and memory at the routers beyond those needed to simply forward the packet. Second, in general, the execution of an active packet at a router may cause more than one active packet to be transmitted from the router. Such behavior is useful, since it allows a packet to fan out across the network, but it is potentially dangerous since it can lead to an exponential growth in the resources used by a single initial packet. Experience with active packet-based systems [9, 8, 23, 22, 24] suggests that denial of service is the single biggest obstacle which must be overcome before such systems are feasible.The proposed research tackles this problem along various fronts. First, the researchers propose to design packet programming languages that make some types of behavior intrinsically impossible. For example, in PLAN [9], packet programs are guaranteed to terminate and thus can never use an un-bounded number of router cycles. The researchers will explore tradeoffs between restricting behavior in terms of resource requirements and limiting the expressibility and thus the flexibility of active packets. However, not all potentially harmful behaviors can be eliminated in this manner. Thus, on a second front, the researchers will consider mechanisms that explicitly account for a packet's resource usage in the network. For example, each packet may carry a resource bound, which is decremented as resources are used, and which triggers termination when the bound is used up. The proposed research combines both implicit and explicit mechanisms for controlling resource usage, with algorithms to control the flow of traffic into the network to decrease the likelihood of denial of service. More generally, one can envisage assessing costs to active packets that execute on congested resources. Thus, on a third front, the researchers propose to investigate mechanisms based on congestion costs to achieve more efficient resource allocations and how they can be facilitated via active packets.Three methodologies will be used to validate proposed solutions. First, the researchers will draw on mathematical modeling to motivate the benefits and investigate the characteristics of the proposed solutions. Second, the researchers will leverage expertise and past work on implementing active networks to demonstrate what is feasible to build, and explore the constraints each solution will place on eventual applications. Finally, the researchers will use network simulation to investigate systems on a scale not achievable on the experimental testbeds.