Kuang-Yi Li

Path layout planning and software based fast failure detection in survivable OpenFlow networks

In an OpenFlow network, the controller is responsible to control and manage the whole network. Although such central control paradigm is easier to achieve efficient network resources usage, the controller becomes the bottleneck when the network size is large. In this paper, we propose using pre-provisioned paths to mitigate the burden on a controller. The basic idea is to setup some background paths in the network. When an ordinary flow coming into the network, the controller only needs to setups flow entries at the ingress and egress switches. An admitted flow follows a preplanned path toward its destination. Except the ingress and egress switches, the flow entries in the remaining switches on the path kept unchanged. Since routing paths for ordinary flows are pre-planned, the whole admission and connection setup process are simplified. Jointly considering load balancing, network survivability, fast failure recovery, and flow entry consumption are the major challenges in this problem. The contribution of this paper includes two parts. First, we propose a path preplanning scheme that takes advantage of interface specific forwarding (ISF) to resolve the problem. We formulate the path preplanning problem as an integer linear programming problem. Optimization based heuristic algorithms are proposed to obtain a near optimal solution. In the second contribution, we propose a software based failure detection and failure location identification scheme. Since some existing switches take longer time to detect a network failure, a reliable lightweight software based failure detection scheme that can fast pinpoint the failure location is needed. We have performed simulations and experiments to evaluate the performance of the proposed schemes. The experimental results indicate that the proposed software failure detection scheme generates very limited overhead on the OpenFlow controller. It can be used to achieve fast failure recovery even when the network is in heavy load state. The simulation results also show that the proposed ISF based path preplanning scheme can reduce controller load and is able to achieve 100% survivability against any single link failure with low bandwidth consumption.

Design and Implementation of a GPON-Based Virtual OpenFlow-Enabled SDN Switch

Passive optical networks (PON) have become a promising solution for accessing networks because of the advantages they offer, such as high efficiency, security, and cost reduction. However, network management in PON is not yet automated and needs network operator intervention. In recent years, software-defined networking (SDN) has become an emerging technology. Through the separation of control and data plane in SDN switches, SDN provides dynamically fine-grained traffic control that enhances total network controllability and manageability. In this paper, we leverage the benefits of gigabit-capable passive optical network (GPON), while enhancing its capabilities on traffic management to the same level as an SDN switch. More specifically, we abstract the underlying physical GPON into an OpenFlow-enabled virtual SDN switch. The virtual switch can be used to connect multiple sites in widespread geographic locations. Similar to a real OpenFlow switch, a GPON virtual switch can be controlled by a standard OpenFlow controller. In our design, an embedded OpenFlow agent resides in the optical line termination (OLT) of the underlying GPON. The agent communicates with the external OpenFlow controller and simultaneously uses optical network unit management and control interface inside the OLT to manage ONUs. We created a prototype system based on a commodity GPON network. In the virtual switch, we implemented all the OpenFlow functions, including packet forwarding, bandwidth metering, statistical data collection, and status reporting. The experimental results show that the GPON virtual switch can correctly perform all the functions defined in the OpenFlow 1.3 specification. Its performance on flow entry modification time, dynamic bandwidth control, and switch status monitoring are comparable to the performance of a real OpenFlow switch.

Design of bandwidth guaranteed OpenFlow virtual networks using robust optimization

In this paper, we address the OpenFlow virtual network (VN) design problem. Unlike most of the existing approaches that directly performing bandwidth slicing on the physical networks, we take traffic uncertainly and statistical multiplexing into consideration. In our system, a user can specify the desired VN topology and the capacity for each virtual link (VL). The bandwidth descriptor for a VL consists of a pair of values for guaranteed bandwidth and upper limit bandwidth. Each user is guaranteed to own the guaranteed bandwidth at any moment without interferences from the bandwidth usages of the other users. Moreover, a user can successfully uses bandwidth up to its upper limit in short time with packet loss probability no more than a pre-defined value. This feature enables OpenFlow networks to provide QoS for network services with bursty traffic. We model this problem as a robust optimization program to jointly determine admission control for VN and routing for VL. We further present an approach for implementing the proposed design using commercial available OpenFlow switches. The performance of the proposed approach is evaluated through simulations and experiments. The results indicate that the proposed design can successfully tolerate traffic uncertainly in OpenFlow networks. Since the proposed approach can use bandwidth with high efficiency, it can admit more VNs than the one without taking statistical multiplexing into consideration.

Software-based fast failure recovery for resilient OpenFlow networks

In this paper, we tackle the failure detection problem in OpenFlow networks. We have examined several commercial OpenFlow switches and found that the long failure detection time makes an OpenFlow network unable to realize fast failure recovery. To resolve this problem, we propose a software based failure detection and fault location identification scheme. Our design can be applied to both in-band controlled and out-of-band controlled OpenFlow networks. In the network, the OpenFlow controller periodically uses monitoring packets to probe network status. Through provisioning monitoring cycles to cover all links in a given network, the OpenFlow controller can detect and pinpoint a failed link within a short time. An algorithm based on the solution of min-max k-Chinese postman problem is proposed to determine the routing of the monitoring cycles. For an in-band controlled network, besides the monitoring cycles, a survivable in-band control tree (ICT) is provisioned to protect in-band control channels. We have conducted experiments to evaluate the performance of the proposed scheme. The results indicate that our approach generates very limited overhead on the OpenFlow controller. It enables an OpenFlow network to achieve fast failure recovery even when the network is in a heavy load state.

Design of SDN based large multi-tenant data center networks

In this paper, we propose a cost effective hybrid SDN-Ethernet network to realize a large data center network. The network can support hundreds of thousands of physical machines (PM). More than 64K tenants can share in a data center and each tenant can individually own 4K VLANs. In addition, each tenant user fully owns the right to assign IP addresses and VLAN IDs to their Virtual Machines (VMs). The network can support VM live migration and fast failure recovery. To enhance survivability, dual SDN controllers are used to manage the network through in-band control. We apply MPTCP to provide multiple concurrent sub-flows to enhance throughput of TCP connections between VMs. A novel congestion aware routing is proposed for network load balancing. We have conducted experiments and simulations to evaluate the performance of the proposed network. The experimental results show that the system can resume connections between VMs in a short time after VM migrations. As a network failure occurs, the proposed failure recovery scheme can restore network connection fast even if the failure disrupting both control and data channels. Our experimental results also reveal that the proposed light weight congestion aware MPTCP routing outperforms the existing ECMP-based and VLB-based routing in two well-known data center topologies.

Optimal bandwidth guaranteed routing and time slot assignment for broadband PLC access networks

High infrastructure availability, easy installation, and low device cost make PLC become one of the promising technologies for broadband communication. In PLC based access network, relay nodes are deployed to extend network coverage and enhance communication quality. In this paper, we consider the bandwidth guaranteed routing and time slot assignment problem in multi-hop PLC networks. Spatial reuse is considered in our system so multiple communication pairs can work at the same time if they do not interfere each other. In particular, our design takes channel quality into consideration in determining the optimal routing. Depending on the receiver node, the sender node uses the best modulation scheme to maximize transmission rate. We formulate the problem as an integer linear programming (ILP) problem. The objective is to minimize the time usage for total data delivery. The output includes optimal routing and collision free slot assignment for each node. To reduce the computation time, we propose a multi-section based algorithm to resolve the problem. We have carried out extensive numerical experiments on several network topologies. Numerical results reveal that the proposed approach obtains significant performance improvement on network resource usage compared to those without taking spatial reuse, routing, and rate adaptation into account.

Energy efficient multi-topology routing configurations for fast failure reroute in IP networks

IETF has defined IP fast reroute (IPFRR) schemes to enable IP networks with fast failover capability. Fundamentally, failure recovery is achieved by provisioning backup paths on standby network resources to provide traffic rerouting for bypassing a failed device. Unlike survivable network designs, the principle used in designing a green network is the opposite. To minimize energy consumption, traffic engineering techniques are applied to aggregate traffic flows on limited network resources. Thus, the unused network devices can be turned off or put on sleep mode to save power. How to reduce power consumption while maintaining high survivability is quite challenging. In this paper, we first evaluate the required number of active links for an IP network operating at its maximum survivability ratio. The results indicate that power saving is quite limited using standard IPFRR. We identify that this result comes from the inflexibility of IP routing using only one set of link metrics. To resolve this issue, we introduced RFC 4915 MTR into the design of a novel green IP network. The method we proposed uses only two routing configurations. Configuration 1 is used for routing traffic in the normal state and Configuration 2 is used to compensate Configuration 1 to enhance network survivability. We have formulated the problem as an integer linear programming problem and propose an optimization-based approach to obtain a solution. The numerical results indicate that the proposed method is not only energy efficient but also can provide 100% network survivability against any single link failure.

Adaptive state transition control for energy-efficient gigabit-capable passive optical networks

In this paper, we investigate power management problems that affect gigabit-capable passive optical networks (GPONs). In GPONs, the basic principle for power reduction is to keep optical network units (ONUs) in the Power Saving state, wherein some of the hardware and software functions are turned off. Current research focuses on scheduling and determining the length of the sleep periods for ONUs that are in the Power Saving state. Our investigation indicates that keeping ONUs in the Power Saving state is not necessarily the most energy-efficient practice. The Power Saving state and the Full Power state must be jointly considered. Our study also reveals that traffic distribution is a critical factor. Considering only the average is insufficient. The variance of packet arrival also must be included when designing a green GPON. We have analyzed the power consumption in a GPON and determined the optimal load threshold for triggering a state transition from the Power Saving state to the Full Power state. For the reverse direction, we propose a neural network-based adaptive control scheme to achieve near optimal control of the transition from Full Power to Power Saving. We also propose a burst transmission scheme to determine the sleep period for an ONU in the Power Saving state. Unlike the proposal of ITU-T, which uses a fixed length for the sleep period, the state sojourn time in our approach is dynamically adjusted. We have carried out extensive simulations to evaluate the performance of the proposed scheme. Simulation results show that the total energy consumption of the proposed scheme is almost equal to the optimal control scheme.

Survivable green active topology design and link weight assignment for IP networks with NotVia fast failure reroute

Survivability and energy efficiency are in conflict with each other on network resources usage in IP networks. The basic principle to achieve high survivability is via provisioning backup paths over spare capacity. However, in designing an energy efficient network, the target is the opposite. In order to minimize energy consumption, routing and traffic engineering techniques are applied to aggregate flows so as to minimize network resource usage. The unused network devices are turned off for power saving. How to reduce power consumption while maintaining high survivability is quite a challenging task. In this paper, we first evaluate the survivability for an IP network using various IP fast reroute (IPFRR) schemes. The results indicate that NotVia IPFRR is one of the most suitable schemes for designing a survivable green IP network. Therefore, in this work, we focus on studying the active topology design and link weight assignment problem for energy efficient IP networks with NotVia IPFRR. We have formulated the problem as an integer linear programming problem and propose a Lagrangean relaxation-based approach to obtain a near optimal solution. The numerical results indicate that the proposed method is not only energy efficient but also has a 100% network survivability capability against any single node failure.

Dynamic load balanced routing in IP networks

In this paper, we address the dynamic load balanced routing problem in IP networks. In current IP networks, shortest paths are selected for routing based on static link weight metrics. Due to the lack of stringent synchronization among routers, dynamically changing link weight would generate transient loops. How to perform dynamic load balanced routing without generating traffic loops become a challenging task. In this work, we propose two approaches, Local Traffic Rerouting (LTR) and Global Traffic Rerouting (GTR), to resolve the problem. Both approaches use a set of static link weight metrics. Each router obtains the metrics through standard link state routing protocol. Based on the link metrics, in each router, a directed acyclic graph (DAG) for each destination node is derived. Taking the advantage of the property that there is no loop in a DAG, each router can make its decision to dynamically select next hop node for routing. The major difference between LTR and GTR is that the former uses only local information while the latter uses global information on routing decision. In LTR, each router monitors its links to obtain the bandwidth utilization. In GTR, the local link utilization information is broadcast to the whole network through routing protocol so each node in the network knows the state of the whole network. We have conducted simulations and experiments to make performance comparisons among various routing approaches. The results indicate that LTR and GTR significantly outperform conventional static routing on minimizing link utilization of the most congested link. They also have performance close to a per flow based centralized controlled routing that is served for a lower bound for performance evaluations.