Srinivasan Seetharaman

Carving research slices out of your production networks with OpenFlow

1. SLICED PROGRAMMABLE NETWORKS OpenFlow [4] has been demonstrated as a way for researchers to run networking experiments in their production network. Last year, we demonstrated how an OpenFlow controller running on NOX [3] could move VMs seamlessly around an OpenFlow network [1]. While OpenFlow has potential [2] to open control of the network, only one researcher can innovate on the network at a time. What is required is a way to divide, or slice, network resources so that researchers and network administrators can use them in parallel. Network slicing implies that actions in one slice do not negatively affect other slices, even if they share the same underlying physical hardware. A common network slicing technique is VLANs. With VLANs, the administrator partitions the network by switch port and all traffic is mapped to a VLAN by input port or explicit tag. This coarse-grained type of network slicing complicates more interesting experiments such as IP mobility or wireless handover. Here, we demonstrate FlowVisor, a special purpose OpenFlow controller that allows multiple researchers to run experiments safely and independently on the same production OpenFlow network. To motivate FlowVisor’s flexibility, we demonstrate four network slices running in parallel: one slice for the production network and three slices running experimental code (Figure 1). Our demonstration runs on real network hardware deployed on our production network at Stanford and a wide-area test-bed with a mix of wired and wireless technologies.

Maturing of OpenFlow and Software-defined Networking through deployments

Software-defined Networking (SDN) has emerged as a new paradigm of networking that enables network operators, owners, vendors, and even third parties to innovate and create new capabilities at a faster pace. The SDN paradigm shows potential for all domains of use, including data centers, cellular providers, service providers, enterprises, and homes. Over a three-year period, we deployed SDN technology at our campus and at several other campuses nation-wide with the help of partners. These deployments included the first-ever SDN prototype in a lab for a (small) global deployment. The four-phased deployments and demonstration of new networking capabilities enabled by SDN played an important role in maturing SDN and its ecosystem. We share our experiences and lessons learned that have to do with demonstration of SDNs potential; its influence on successive versions of OpenFlow specification; evolution of SDN architecture; performance of SDN and various components; and growing the ecosystem.

The Stanford OpenRoads deployment

We have built and deployed OpenRoads [11], a testbed that allows multiple network experiments to be conducted concurrently in a production network. For example, multiple routing protocols, mobility managers and network access controllers can run simultaneously in the same network. In this paper, we describe and discuss our deployment of the testbed at Stanford University. We focus on the challenges we faced deploying in a production network, and the tools we built to overcome these challenges. Our goal is to gain enough experience for other groups to deploy OpenRoads in their campus network.

Characterizing and Mitigating Inter-domain Policy Violations in Overlay Routes

The Internet is a complex structure arising from the interconnection of numerous autonomous systems (AS), each exercising its own administrative policies to reflect the commercial agreements behind the interconnection. However, routing in service overlay networks is quite capable of violating these policies to its advantage. To prevent these violations, we see an impending drive in the current Internet to detect and filter overlay traffic. In this paper, we first present results from a case study overlay network, constructed on top of Planetlab, that helps us gain insights into the frequency and characteristics of the different inter-domain policy violations. We further investigate the impact of two types of overlay traffic filtering that aim to prevent these routing policy violations: blind filtering and policy- aware filtering. We show that such filtering can be detrimental to the performance of overlay routing. We next consider two approaches that allow the overlay network to realize the full advantage of overlay routing in this context. In the first approach, overlay nodes are added so that good overlay paths do not represent inter-domain policy violations. In the second approach, the overlay acquires transit permits from certain ASes that allow certain policy violations to occur. We develop a single cost-sharing framework that allows the incorporation of both approaches into a single strategy. We formulate and solve an optimization problem that aims to determine how the overlay network should allocate a given budget between paying for additional overlay nodes and paying for transit permits to ASes. We illustrate the use of this approach on our case study overlay network and evaluate its performance under varying network characteristics.

Preemptive Strategies to Improve Routing Performance of Native and Overlay Layers

Overlay routing is known to cause undesired instability in a network by operating in a selfish manner. The objectives of overlay routing, such as optimizing end-to-end latency, are often in conflict with the objectives of traffic engineering in the native layer, which is concerned about balancing load. In our work, we build on past research that has investigated the recurring non-cooperative interaction between overlay routing and traffic engineering, and develop strategies that improve the routing performance of a particular layer with incomplete information about the other layer. In our strategies, one layer acts as a leader that predicts the followers reaction and undertakes countermeasures to prevent future deterioration in performance. Specifically, we propose two classes of strategies - friendly or hostile - for each layer. By simulating under different network characteristics, we show that these preemptive strategies achieve near-optimal performance for the leader and increase the overall stability of the network. Furthermore, we observe that the best performance for a particular layer is achieved only when the goals of the other layer are completely violated, thereby motivating a higher level of selfishness.

On the Interaction Between Dynamic Routing in Native and Overlay Layers

Overlay networks have recently gained attention as a viable alternative to overcome functionality limitations of the Internet. We are concerned with scenarios where a dynamic routing protocol is employed in the overlay network to adapt overlay routing tables to changing network conditions. At the same time, the native network over which the overlay is built also runs its own set of dynamic routing protocols. We are interested in investigating the behavior of this mixed routing environment and in particular the characteristics of the interaction between these two routing layers. In this paper, we focus on the specific problem of rerouting around failed links. We first study a Dual Rerouting scenario in which the two routing layers run completely independent of each other. Our goal is to understand the effect of the various settings of routing protocol parameters on the packet loss, number of route flaps, and the optimality of the adopted overlay path. We show that Dual Rerouting provides relatively fast path recovery. But, it tends to be sub-optimal in terms of the number of route flaps and the overlay path cost inflation. This is due to the overlap of functionality between the two layers, unawareness of the other layer’s decisions, and lack of flexibility. We next investigate schemes that increase awareness of the native routing protocol and its parameters at the overlay layer. We consider three such approaches: Probabilistically Suppressed Overlay Rerouting, Deferred Overlay Rerouting and Follow-on Suppressed Overlay Rerouting. We show that with such schemes one can trade off longer path recovery times with improvements in route flapping and path cost inflation. However, there is a fundamental limit on the amount of achievable gain if we receive no support from the native layer. To counter that, we propose a novel approach towards the tuning of native layer parameters to suit the functioning of the overlay layer.

Multi-layer Monitoring of Overlay Networks

Monitoring end-to-end paths in an overlay network is essential for evaluating end-system performance and for troubleshooting anomalous behavior. However, conducting measurements between all pairs of overlay nodes can be cumbersome and expensive, especially in a large network. In this paper, we take a different approach and explore an additional degree of freedom, namely, monitoring native links. We allow native link measurements, as well as end-to-end overlay measurements, in order to minimize the total cost of monitoring the network. We formulate an optimization problem that, when solved, identifies the optimal set of native and overlay links to monitor, and a feasible sequence of arithmetic operations to perform for inferring characteristics of the overlay links that are not monitored directly. We use simulations to investigate how various topological properties may affect the best monitoring strategy. We also conduct measurements over the PlanetLab network to quantify the accuracy of different monitoring strategies.

Resolving cross-layer conflict between overlay routing and traffic engineering

Overlay routing is known to cause undesired instability in a network by operating in a selfish manner. The objectives of overlay routing, such as optimizing end-to-end latency, are often in conflict with the objectives of traffic engineering (TE) in the native layer, which is concerned about balancing load. In our paper, we build on past research that has investigated the recurring noncooperative interaction between overlay routing and traffic engineering, and develop strategies that improve the routing performance of a particular layer with incomplete information about the other layer. In our strategies, one layer acts as a leader that predicts the followers reaction and undertakes countermeasures to prevent future deterioration in performance. Specifically, we propose two classes of strategies--friendly or hostile--for each layer. By simulating under different network characteristics, we show that these preemptive strategies achieve near-optimal performance for the leader and increase the overall stability of the network. Furthermore, we observe that the best performance for a particular layer is achieved only when the goals of the other layer are completely violated, thereby motivating a higher level of selfishness.

Managing inter-domain traffic in the presence of bittorrent file-sharing

Overlay routing operating in a selfish manner is known to cause undesired instability when it interacts with native layer routing. We observe similar selfish behavior with the BitTorrent protocol, where its performance-awareness causes it to constantly alter the routing decisions (peer and piece selection). This causes fluctuations in the load experienced by the underlying native network. By using real BitTorrent traces and a comprehensive simulation with different network characteristics, we show that BitTorrent systems easily disrupt the load balance across inter-domain links. Further, we find that existing native layer traffic management schemes suffer from several downsides and are not conducive to deployment. To resolve this dilemma, we propose two BitTorrent strategies that are effective in resolving the cross-layer conflict.

Inter-domain policy violations in multi-hop overlay routes: Analysis and mitigation

The Internet is a complex structure arising from the interconnection of numerous autonomous systems (AS), each exercising its own administrative policies to reflect the commercial agreements behind the interconnection. However, routing in service overlay networks is quite capable of violating these policies to its advantage. To prevent these violations, we see an impending drive in the current Internet to detect and filter overlay traffic. In this paper, we first present results from a case study overlay network, constructed on top of PlanetLab, that helps us gain insights into the frequency and characteristics of the different inter-domain policy violations. Further, we investigate the impact of two types of overlay traffic filtering that aim to prevent these routing policy violations: blind filtering and policy-aware filtering. We show that such filtering can be detrimental to the performance of overlay routing. We next consider two approaches that allow the overlay network to realize the full advantage of overlay routing in this context. In the first approach, overlay nodes are added so that good overlay paths do not represent inter-domain policy violations. In the second approach, the overlay acquires permits from certain ASes that allow certain policy violations to occur. We develop a single cost-sharing framework that allows the incorporation of both approaches into a single strategy. We formulate and solve an optimization problem that aims to determine how the overlay network should allocate a given budget between paying for additional overlay nodes and paying for permits (transit and exit) to ASes. We illustrate the use of this approach on our case study overlay network and evaluate its performance under varying network characteristics.