Ruediger Martin

Network resilience through multi-topology routing

In this paper we propose the use of multi-topology (MT) routing for network resilience against link and node failures. We describe the multi-topologies by an n-dimensional vector of different link costs for all links in the network. It is the base for the calculation of n shortest path trees from any node to all other destinations, i.e. for n virtual routing topologies. We define the link costs in such a way that the routing topologies complement each other in the sense that at least one valid route remains in a single link or node failure scenario for each pair of nodes in at least one routing topology. In such a failure case, packets are rather deviated over the intact routing topology than discarded. The recovery speed of the presented mechanism is very fast and can be compared to fast rerouting mechanisms in MPLS, which reduce packet drops to a minimum. In contrast to MPLS, MT routing is still a pure IP-based solution that retains the scalability and the robustness of IP routing.

Resilience Analysis of Packet-Switched Communication Networks

In the presence of local network outages, restoration and protection switching mechanisms redirect the traffic over alternative paths to mitigate the effect of failures. However, some failure combinations still lead to loss of ingress-egress connectivity within a network or to severe congestion due to rerouted traffic. Congestion may also be caused by unexpected traffic shifts due to changed user behavior or due to changes of interdomain routing. This paper presents a framework for the analysis of ingress-egress unavailability and link congestion due to: 1) failures; 2) changes of user behavior; and 3) changed interdomain routing. It proposes algorithms to find the most probable combinations of 1)-3) according to some models, and to evaluate the connectivity and the relative link load of the network under these conditions. We have implemented this concept in a software tool and its visualization of the results leads to a comprehensive view of the networks resilience. It helps to anticipate potential ingress-egress disconnection and congestion before failures and overload occur or before planned modifications (new infrastructure, new routing, new customers) take effect. Thus, it detects weak points in a network, predicts the effectiveness of potential upgrades, and thereby supports careful bandwidth overprovisioning.

Accuracy and Dynamics of Hash-Based Load Balancing Algorithms for Multipath Internet Routing

This paper studies load balancing for multipath Internet routing. We focus on hash-based load balancing algorithms that work on the flow level to avoid packet reordering which is detrimental for the throughput of transport layer protocols like TCP. We propose a classification of hash-based load balancing algorithms, review existing ones and suggest new ones. Dynamic algorithms can actively react to load imbalances which causes route changes for some flows and thereby again packet reordering. Therefore, we investigate the load balancing accuracy and flow reassignment rate of load balancing algorithms. Our exhaustive simulation experiments show that these performance measures depend significantly on the traffic properties and on the algorithms themselves. As a consequence, our results should be taken into account for the application of load balancing in practice.

Network Dimensioning for the Self-Protecting Multipath: A Performance Study

The self-protecting multipath (SPM) is a simple protection switching mechanism that can be implemented, e.g., by MPLS. We present a linear program to optimize the SPM load balancing parameters for network dimensioning. Our study shows that the SPM is a very efficient mechanism in the sense that it requires only little backup capacity since it outperforms the p-cycle approach and the shortest path rerouting by far. The investigation of the computation time and the memory consumption recommends the Simplex method as an LP solver rather than an interior point method (IPM). The computation time of the program depends mainly on the number of links in the network and it is well feasible for small and and medium size networks. For large networks, however, fast heuristics are required.

Comparison of border-to-border budget based network admission control and capacity overprovisioning

There are two basic approaches to achieve Quality of Service (QoS) for communication networks: admission control (AC) and capacity overprovisioning (CO). CO is simple and cheaper to implement than AC but AC requires less capacity to fulfill QoS criteria since overload traffic can be blocked. There is an almost religious war between scientists working on both concepts. In this paper we try to contribute insights for this discussion by quantifying the capacity savings potential of AC under various networking conditions.

Capacity Requirements for the One-to-One Backup Option in MPLS Fast Reroute

MPLS fast reroute (MPLS-FRR) mechanisms deviate the traffic in case of network failures at the router closest to the outage location to achieve an extremely fast reaction time. We review the one-to-one backup and the facility backup that are options for MPLS-FRR to deviate the traffic via a detour or a bypass around the failed elements, respectively. Basically, the backup paths can take the shortest path that avoids the outage location from the point of local repair to the tail-end router or to the merge point with the primary path. We suggest two simple modifications that lead to a new path layout which can be implemented by one-to-one and by facility backup. We evaluate the backup capacity requirements, the length of the backup paths, and the number of backup paths per primary path in a parametric study regarding the network characteristics. Our proposals save a considerable amount of backup capacity compared to the standard mechanisms. They are suitable for application in practice since they are simple and conform to the standards.

Service differentiation with MEDF scheduling in TCP/IP networks

The Differentiated Services architecture implements appropriate Per Hop Behaviors (PHBs) for service differentiation to achieve Quality of Service (QoS) in Next Generation Networks (NGNs). In this paper, we review several buffer management and scheduling algorithms that may be applied in this context. The major part of this work focuses on the interaction between TCP and the Modified Earliest Deadline First (MEDF) scheduling algorithm. It is an appealing option for service differentiation because it relies on a single parameter only and because this parameter can be configured independently of the current share of high priority traffic in the network. This is contrary to many other scheduling disciplines or buffer management strategies. The unavailability of this information in todays Internet makes MEDF an interesting candidate for commercial application.

Risk assessment of end-to-end disconnection in IP networks due to network failures

Restoration and protection switching mechanisms in IP networks are triggered by link or node failures to redirect traffic over backup paths. These mechanisms are no longer effective if a network becomes disconnected after a failure. The risk of end-to-end disconnection increases if the nodes of a network are only sparsely meshed or if multiple network failures occur simultaneously. This leads inevitably to violations of service level agreements with customers and peering network providers. In this paper, we present a method to assess the risk of end-to-end disconnection in IP networks due to network failures. We calculate the disconnection probabilities for all pairs of network nodes taking into account a set of probable network failures. The results are considered from different perspectives. This helps to identify weak spots of the network and to appropriately upgrade its topological infrastructure with additional links. We implemented the concept in a software tool which assists network providers with assessing the risk of disconnection in their network prior to any network failure and to take appropriate actions.

Performance of TCP/IP with MEDF scheduling

To achieve Quality of Service (QoS) in Next Generation Networks (NGNs), the Differentiated Services architecture implements appropriate Per Hop Behavior (PHB) for service differentiation. Common recommendations to enforce appropriate PHB include Weighted Round Robin (WRR), Deficit Round-Robin (DRR) and similar algorithms. They assign a fixed bandwidth share to Transport Service Classes (TSCs) of different priority. This is a viable approach if the ratio of high priority traffic TSC high over low priority traffic TSC low is known in advance. If TSC high holds more and TSC low less users than expected, the QoS for TSC high can be worse than for TSC low . As shown in preceding work, the Modified Earliest Deadline First (MEDF) algorithm heals this problem on the packet level. Therefore, we investigate its impact in congested TCP/IP networks by simulations and show its attractiveness as a powerful service differentiation mechanism.