András Kern

Scalable fault management for OpenFlow

In the OpenFlow based split architecture, data-plane forwarding is separated from control and management functions. Forwarding elements make only simple forwarding decisions based on flow table entries populated by the controller. While OpenFlow does not specify how topology monitoring is performed, the centralized controller can use Link-Layer Discovery Protocol (LLDP) messages to discover link and node failures and trigger restoration actions. This monitoring and recovery model has serious scalability limitations because the controller has to be involved in the processing of all of the LLDP monitoring messages. For fast recovery, monitoring messages must be sent with millisecond interval over each link in the network. This poses a significant load on the controller. In this paper we propose to implement a monitoring function on OpenFlow switches, which can emit monitoring messages without posing a processing load on the controller. We describe how the OpenFlow 1.1 protocol should be extended to support the monitoring function. Our experimental results show that data plane fault recovery can be achieved in a scalable way within 50 milliseconds using this function.

Surviving multiple network failures using shared backup path protection

This paper studies the possibilities of protecting backbone networks from multiple simultaneous failures. Although methods that prepare the networks to survive single failures were extensively studied in the literature, in order to provide extra high availability for mission-critical applications, the possibility of more than one simultaneous failure has to be considered too. The general technique of protecting a traffic flow is to establish a backup path where the traffic is redirected when a failure occurs along its active path. Using this technique the total capacity reservation is at least doubled compared to the non-protected case. To avoid this large amount of bandwidth reservation by the backup paths, the shared backup path protection paradigm can be used, which is a promising scheme favored by the Internet engineering task force. In this paper the two backup path based protection scheme is introduced, which provides survivability against any two simultaneous failures. Further, a novel shared protection strategy is proposed that improves the failure tolerance of the traditional single backup path based protection. The performance of the different protection types is examined through simulations.

Bandwidth guarantees for resilient Ethernet networks through RSTP port cost optimization

Ethernet protocol is the most wide-spread protocol in the local area network (LAN) environment. It is cost effective, simple, and provides high speeds, exactly what is needed in the provider network. However, deployment in the provider network imposes carrier-grade requirements. Standardization bodies realized this, and they are extending its capabilities: QoS support and traffic management by 802.1Q virtual LANs (VLANs), traffic engineering by 802.1s multiple spanning tree protocol (MSTP), administration and management issues by 802.1 ag connectivity fault management (CFM). However, with carrier-grade extensions Ethernet started to lose an important property: simplicity. Often the topology of the Ethernet aggregation is simple, tree-like, so complex protocols like MSTP are not required even when resilience is supported by adding several links. In most cases the plain rapid spanning tree protocol (RSTP) can provide the necessary restoration and loop protection capability. However, the default settings of the RSTP protocol may not provide optimal network utilization, and it is difficult to predict its behavior after a failure. In this paper we present a tool for RSTP optimization, which engineers the network for best utilization while also ensures that optimal paths will be selected after a link failure. The result of the optimization is a port cost set for all bridge interfaces. The optimization is performed off-line using an integer linear program (ILP). We show that compared to the default settings higher throughput can be achieved, while the required bandwidth is guaranteed in case of any link failure.

Optimizing QoS aware Ethernet spanning trees

Ethernet is gaining in importance in both access and metro networks. As a layer 2 technology, Ethernet gives a basic framework for routing, QoS and traffic engineering (TE), as well as a protocol for building up trees. IEEE 802.1 standards define default configuration parameters considering the topology only. We propose methods for resource management in Ethernet networks through spanning tree optimization for both STP (spanning-tree protocol) (IEEE 802.1D) and MSTP (multiple spanning-tree protocol) (IEEE 802.1s). As a result of optimization, we assign costs to the bridge ports in the network to build trees based on these costs via STP and MSTP. These trees yield optimized routing, TE and support for different QoS classes. We show on typical metro-access networks that, through optimization, the total network throughput can be significantly increased for both enforcing fairness or allowing starvation of some demands. This gain can be realized by simultaneously assigning demands to trees and routing these trees.

Splitarchitecture: SDN for the carrier domain

The concept of SDN has emerged as a way to address numerous challenges with traditional network architectures by decoupling network control and forwarding. So far, the use cases for SDN mainly targeted data-center applications. This article considers SDN for network carriers, facing operation of large-scale networks with millions of customers, multiple technologies, and high availability demands. With specific carrier-grade features such as scalability, reliability, flexibility, modularity, and virtualization in mind, the SPARC EU project has developed the SPARC SplitArchitecture concept. The SplitArchitecture control plane allows hierarchical layering of several control plane layers which can be flexibly mapped to data plane layers. Between control layers open interfaces are used. Furthermore, SplitArchitecture proposes an additional split of forwarding and processing functions in data path elements, enabling switch based OAM functionality and handling of tunneling techniques. The SplitArchitecture concept is evaluated in a prototype demonstrating an SDN version of BRAS: the floating BRAS. The floating BRAS allows creation of residential Internet access services with dynamically managed BRAS instances. The demo is based on a controller enabling protected MPLS transport connections spanning SDN-controlled aggregation and IP/MPLS-based core domains. The prototype showcases how SplitArchitecture enables virtualization of service nodes in an SDN-controlled network, leading to increased flexibility in configuration and placement of service creation functions. Overall, the results show that it is technically and economically beneficial to apply SDN, and specifically the SplitArchitecture concept, to the carrier domain.

/spl lambda/-path fragmentation and de-fragmentation through dynamic grooming

In multi-layer networks, where more than one layer is dynamic, i.e., connections are set up using not only the upper, e.g., IP layer but the underlying wavelength layer as well leads often to suboptimal performance. In this paper we discuss the lightpath fragmentation and de-fragmentation problem, where the lightpath system cannot follow the traffic changes fast enough. We show what it depends on, when is it critical and how much does it deteriorate the performance of the network.

Optimized QoS protection of Ethernet trees

Ethernet is being increasingly employed in metro networks. Ethernet gives routing schemes and protocols for building up trees, for instance STP, RSTP and MSTP. It also implies simple restoration mechanisms. In this paper we propose an optimization framework, where the Ethernet MSTP trees are protected and QoS is guaranteed even after a failure. The optimization is based not only on the topology, but it also takes traffic conditions and QoS constraints into account. The numerical results show that the proposed optimization significantly increases the throughput of the network. The best result can be achieved when preemption is assumed, i.e., when the best effort traffic may remain unprotected, but not the high priority one. This way high throughput can be realized at normal operation, while it still protects prioritized traffic in case of a failure. Furthermore, protection mechanisms act faster than the standard restoration mechanism resulting shorter out-of-service times, and therefore higher availability.

Multi-layer traffic engineering through adaptive λ-path fragmentation and de-fragmentation

In Multi-Layer networks, where more than one layer is dynamic, i.e., connections are set up using not only the upper, e.g., IP layer but the underlying wavelength layer as well leads often to suboptimal performance due to long wavelength paths, that do not allow routing the traffic along the shortest path. The role of MLTE (Multi-Layer Traffic Engineering) is to cut these long wavelength paths into parts (fragments) that allow better routing at the upper layer (fragmentation), or to concatenate two or more fragments into longer paths (defragmentation) when the network load is low and therefore less hops are preferred. In this paper we present a new model (GG: Grooming Graph) and an algorithm for this model that supports Fragmentation and De-Fragmentation of wavelength paths making the network always instantly adapt to changing traffic conditions. We introduce the notion of shadow capacities to model “lightpath tailoring”. We implicitly assume that the wavelength paths carry such, e.g., IP traffic that can be interrupted for a few microseconds and that even allows minor packet reordering. To show the superior performance of our approach in various network and traffic conditions we have carried out an intensive simulation study.

Scalable Tree Optimization for QoS Ethernet

The low cost and the wide availability make Ethernet the dominant networking technique of access networks. Standardization further extends the capabilities of Ethernet by adding traffic separation and prioritization (802.1Q) and by allowing use of multiple spanning trees (802.1s). In this paper we present an Ethernet based QoS architecture for broadband services with triple play support. We propose an optimization framework with Traffic Engineering where the MSTP trees are spanned taking both the traffic conditions and the QoS requirements into account. The optimization is based on our previously presented formal model. Here we propose a novel decomposition based tree constructing heuristic with high scalability. We conduct simulations to evaluate the performance of the heuristic method that provides nearly optimal solutions within acceptable time constraints.

Scalability aspects of centralized control of MPLS access/aggregation network

We apply the concept of centralized control to the MPLS access/aggregation network segment of service provider networks. We describe a numerical model to capture the scale and scope of the access/aggregation network and explain the corresponding scalability issues. We compare this numerical model to the experimental results obtained on a test-bed with OpenFlow 1.1 compatible central controller and forwarding elements. We show that the central controller implementation we have mostly matches the performance requirements of the access/aggregation networks for different scenarios.