Martin Horneffer

Multi-layer capacity planning for IP-optical networks

We consider a pragmatic multi-layer capacity planning approach for an IP over optical network, which builds on existing planning practices. We first describe a router bypass process that considers the real impact of the IP layer topology change, given the actual behavior of the IP layer, leveraging a commercial IP planning tool. We then consider the design and cost impact of several multi-layer restoration schemes. We compute the savings achieved on two real-world core network models. The resulting savings are similar for both networks and are very promising.

Quality-of-service class specific traffic matrices in ip/mpls networks

In this paper we consider the problem of determining traffic matrices for end-to-end demands in an IP/MPLS network that supports multiple quality of service (QoS) classes. More precisely, we want to determine the set of traffic matrices Ti for each QoS class i separately. Ti contains average bandwidth levels for QoS class i for every pair of routers within the network. We propose a new method for obtaining QoS class specific traffic matrices that combines estimation and measurement methods: We take advantage of the fact that the total traffic matrix can be measured precisely in MPLS networks using either the LDP or RSVP-TE protocol. These measurements can then be used in a mathematical model to improve estimation methods - known as network tomography - that estimate QoS class specific traffic matrices from QoS class specific link utilizations. In addition to the mathematical model, we present results of the proposed method from its application in Deutsche Telekoms global IP/MPLS backbone network and we show that the estimation accuracy (mean relative error) is improved by a factor of 2.5 compared to results from network tomogravity. We investigate the structure of the estimated traffic matrices for the different QoS classes and motivate in this paper why QoS class specific traffic matrices will be essential for efficient network planning and network engineering in the future.

Root causes for iBGP routing anomalies

Today, spreading global routing information within an Autonomous System (AS) is usually implemented by means of internal BGP (iBGP), a particular operational mode of the Border Gateway Protocol (BGP) [1]. Considering the research activities around BGP, anomalies turned out to be one major issue of iBGP: But even if the trouble iBGP anomalies induce in practice is well known for nearly a decade [2], [3], new iBGP extensions still define highly problematic protocol modifications. Apparently, protocol designers are either not aware of the conceptual defects that induce anomalies or validating correctness appears to be too complex. In this paper, we provide a basis to solve both problems. Focussing the basic correctness properties of routing protocols, we will show that iBGP anomalies are caused by five root causes. These causes provide protocol designers with a tool set to reveal problematic properties of a concept and verify correctness. Based on exemplary studies for two common iBGP architectures, we will provide detailed guidance for practical usage.

An inherently anomaly-free iBGP architecture

Especially for large Network Operators (NOs), eliminating routing anomalies is an important aspect for the internal BGP design. To avoid such unwanted effects, classical architectures [1], [2], [3] have need of certain restrictions kept in both, the network design and the iBGP peering. However, these restrictions are in conflict with the optimized network designs NOs seek. In this study we develop an inherently anomaly-free iBGP architecture that takes the demands of NOs into account. It is based on a central server that guarantees consistent local views in the entire system. This is done by exchanging additional routing information with the Border Routers of the Autonomous System. The architecture implements the results of iBGP analyses made in [4]. Despite not being that as flexible as classical information reduction techniques, our design scales equal or better in practice. All required protocol extensions are supported by IETF Internet Drafts from persons affiliated with important router vendors.

Combining LDP Measurements and Estimation Methods for Traffic Matrices in IP/MPLS Networks

In this paper we consider the problem of determining the traffic matrix of end-to-end demands in an IP network that uses multiprotocol label switching (MPLS). Although it is possible to measure end-to-end demands directly in this case using RSVP-TE tunnels it is not always desired to establish a full mesh of tunnels which is necessary to measure the complete traffic matrix: This introduces an additional routing layer and increases the operational complexity of the network. We propose in this paper a framework to generate traffic matrices by combining measurements with estimation techniques. We use statistics of the label distribution protocol (LDP) to obtain end-to-end traffic flows. These label swapping statistics can be used to calculate flows on a high aggregation level and allow for a simple and inexpensive measurement infrastructure. It depends on the network topology as well as vendor equipment what part of the traffic matrix can be determined exactly in this way. We investigate different estimation methods to complete the traffic matrix and introduce a model to additionally estimate the decomposition of the traffic matrix into traffic matrices for different QoS classes

The Scope of the IBGP Routing Anomaly Problem

Correctness problems in the iBGP routing, the de-facto standard to spread global routing information in Autonomous Systems, are a well-known issue. Configurations may route cost-suboptimal, inconsistent, or even behave non-convergent and -deterministic. However, even if a lot of studies have shown many exemplary problematic configurations, the exact scope of the problem is largely unknown: Up to now, it is not clear which problems may appear under which iBGP architectures. The exact scope of the iBGP correctness problem is of high theoretical and practical interest. Knowledge on the resistance of specific architecture schemes against certain anomaly classes and the reasons may help to improve other iBGP schemes. Knowledge on the specific problems of the dierent schemes helps to identify the right scheme for an AS and develop workarounds.

Scalability considerations for an anomaly-free iBGP routing

Scalability is an important aspect when spreading externally learned routing information via the Border Gateway Protocol within Autonomous Systems. However, research showed that common iBGP scalability techniques like Route Reflection and AS Confederations are involved with anomalies in general. These unwanted effects can be inherently avoided by providing routers with individual sets of routing information. Implementing a modified iBGP routing information exchange in productive systems requires a scalability which is comparable to conventional techniques. Our studies show that providing routers with suitable individual information should not cause any serious decline at this point: While the most crucial parameters proved to be not problematic, only the number of paths managed by routers turn out to be hard to predict. Hence, we examine this parameter for the productive system of a large Service Provider, showing only irrelevant increases. Our results indicate that this approach could be an alternative for future iBGP routing architectures.

Scalability of iBGP path diversity concepts

Improving the path diversity seems to be the next fundamental step in the iBGP evolution. Focusing the advantages an improvement of the path diversity implies, network protocol designers have disregarded the most critical drawback so far: The effect on the scalability of the iBGP routing, a fundamental requirement for production usage. This aspect is examined by the analyses discussed in our paper. In this paper, we provide the theoretical groundwork for scalability analyses of four highly relevant path diversity schemes. Based on this groundwork, we exemplarily predict the information load the schemes induce in a system of a large ISP. Generalizing the system-specific results, we give an outlook on the load that can be expected in comparable ASs. We found that for two schemes currently in the standardization process, scalability problems in large ASs as they are operated by ISPs seem likely.

Benefits from 2-Layer Traffic Engineering for IP/MPLS Networks

Multi-layer aspects of network and traffic engineering have been extensively studied from various perspectives and potential benefits for savings in capital and operational expenditures have been demonstrated. Integrated approaches for traffic engineering and network planning come closer towards a practical implementation with the progress in standardization of protocols like generalized multiprotocol label switching (GMPLS). In this paper we consider the aspect of optimizing the routing of IP links in the underlying transport layer in order to improve the failure scenarios in the IP/MPLS layer. The goal is to minimize worst case network utilizations and avoid potential overload situations. We propose an approximation algorithm to solve the general 2-layer optimization problem in large realistic network scenarios and conduct a case study to validate our approach. On the one hand we address a rather small aspect of multi-layer network optimization here but on the other hand we identify benefits that allow for a short-term implementation.