Veena B. Mendiratta, Charles Witschorik (Bell Labs, Lucent Techonolgies, USA):
Survivability of Telecommunications Systems: Concepts and Architectures

Public telecommunications systems have evolved to be highly reliable through hardware and software redundancy, fault recognition and containment, and fast and effective recovery. There is now a new emphasis for telecommunications systems to also be highly survivable so as to limit the extent of service loss after a catastrophic loss of infrastructure. This tutorial will provide an introduction to the subject of telecommunications system survivability from a practitioner's perspective. Our focus is exclusively on the voice network for wireline, cellular and voice over packet implementations.

The following areas are covered. The introduction discusses the motivation for the need for survivable telecommunications systems with examples and related work in this area. The background section covers definitions and requirements, followed by a section on network components for wireline, wireless and packet networks. The section on survivability principles defines the layered telecommunications architecture and the hardware and software components in each layer. The recovery section defines the automatic detection and recovery principles currently used in telecommunications systems to achieve reliability. An overview of the processes and procedures required for effecting recovery after a loss are presented in the next section. Principles for subscriber database recovery and survivability are presented in the database section. The transport section briefly addresses transport survivability issues. The section on alternative architectures brings together the architecture principles described earlier to define alternative architectures for survivability for wireline, cellular and voice over packet networks. The alternative architectures are presented in the context of incremental changes to existing networks to improve survivability as well as in the context of new installations (green-field) which can be closer to an ideal architecture with respect to survivability. The last section summarizes what is required for architecting a survivable voice telecommunications system. A list of references is also included.

Max Walter (TU München, Germany):
Inter-component dependencies in models for dependable systems

Quantitative modeling methods for dependable systems can be divided into two classes: combinatorial methods like fault trees and reliability block diagrams and state-based methods like Markov Chains (MC) and Stochastic Petri Nets (SPN). Only by using the latter class it is possible to take into account dependencies in the failure and repair behavior of a system's components.

This tutorial shows how most common dependencies including limited repair capacities, non-zero failover-times, failures with a common cause, failure propagation, and imperfect coverage can be modeled in practice using MCs and SPNs. It also shows the difficulties and limitations of these modeling methods in terms of manageability and presents some solutions as well as open problems as found in the recent literature.

Jens B. Schmitt and Frank A. Zdarsky (Uni Kaiserslautern, Germany):
Network Calculus - A Life After IntServ

Performance analysis cannot be mentioned without raising associations with queueing theory as its classical main tool. Yet, besides traditional queueing theory a new theory called network calculus has emerged as a valuable tool for analyzing a system's worst-case behavior. It allows to determine such characteristics of data flows as the maximum latency or the minimum bandwidth, as long as bounds can be specified for these flows in the form of so-called arrival curves. Network calculus can be viewed as a system theory of deterministic queuing systems. The basis for this system theory is the min-plus algebra, in which operations '+' and '*' of conventional algebra are substituted by the minimum operation and '+', respectively.

Network calculus was conceived in the mid-nineties in the context of providing Quality of Service in the Internet and served as the mathematical foundation of the Integrated Services (IntServ) architecture. Yet it is far more versatile in its applications, and we believe that it is still at the beginning of its relatively short existence as a theoretical foundation for the performance modeling of distributed systems.

This tutorial on network calculus is organized into three parts: (i) mathematical foundations and traditional applications of network calculus, (ii) advanced aspects of network calculus and extensions towards a system theory for distributed systems, (iii) practical aspects of the implementation of support tools. In particular this last aspect of providing support tools is still very much in its infancy. We will introduce first steps towards a general "Network Calculator".

A brief outline of the tutorial is as follows:

1. Mathematical Foundations and Traditional Applications of Network Calculus
   • min-plus algebra, min-plus convolution / deconvolution
   • arrival and service curves
   • basic theorems of network calculus
   • application to the IntServ architecture, "pay bursts only once"-principle
2. Advanced Aspects of Network Calculus and Extensions Towards a System Theory for Distributed Systems
   • transformations for a min-plus system theory
   • multiplexing of aggregates, application to the DiffServ architecture
   • bounds in feed-forward networks, applications to wireless sensor networks
3. Practical Aspects of the Implementation of Support Tools
   • algorithms for network analysis
   • pragmatic limitation to piecewise-linear functions
   • heuristics for controlling computational complexity
   • introduction of the Network Calculator

Gerhard Haßlinger (T-Systems, Germany):
Peer-to-Peer Networking and Traffic Management on Internet Platforms

Since the millennium, peer-to-peer (P2P) applications became a main driver of increasing Internet traffic. File sharing often contributes 50% - 80% of the traffic volume on developing broadband access platforms in Europe and North America. Although unresolved legal issues still leave an uncertain future for a major part of ongoing P2P activity, a broad spectrum of emerging and already established applications are build on peer-to-peer overlay structures. After Skype launched the currently most popular voice over IP network within the last two years, new P2P approaches evolved e.g. for content distribution, audio/video streaming, online gaming, cooperative work tools etc. A basic P2P network feature is the ability to distribute resources (storage, computation power, bandwidth) required for a service among large user communities, together with scalable adaptation to current demands and with replications as needed. In this way, new services can be established without own server and network infrastructure or in hybrid architectures with server functions being reduced to a minimum while maintaining central control. As a consequence, shifts in the traffic profiles are visible on Internet platforms with regard to

• the size of traffic flows within each TCP connection due to fragmentation of large files by P2P protocols into smaller data chunks to be downloaded from several sources in parallel,
• the burstiness of traffic on Internet links, where a smoothing effect is observed for the traffic rate variability in short time scales as well as in daily profiles,
• the transmission paths, which are determined e.g. by selecting sources for a download.

P2P protocols establish an application layer routing which may differ from the routing on network layer causing unnecessary backbone and "off-net" traffic between network providers. On the other hand, they distribute and thus reduce spikes in local access pattern caused e.g. by spontaneously changing popularity of servers. Geographical location of data sources are often structured in P2P communities separated by language, culture and other social factors and preferences in different European countries. The tutorial starts with a short overview of popular and emerging peer-to-peer protocols. Besides network structures, the initialization and the course of data transfers is addressed with focus on the resulting traffic profiles. Methods for analyzing P2P traffic in IP platforms are summarized and approaches are discussed to cope with overhead or an inefficient assignment between the P2P overlay and the IP network structure, e.g. by the use of caches. Expected smoothing effects are compared to measurement on links in broadband access platforms over different time scales. Even if background data transfers for file sharing do not impose strict demands for quality of service (QoS), the shift in the entire application and traffic mix mainly influences traffic management including appropriate measures to support QoS for an increasing variety of services in IP and next generation networks. The tutorial includes the following main parts:

1. Peer-to-Peer Protocols, Applications and Usage Pattern
2. Cross Layer Issues: IP versus P2P Overlay Network Structures
3. Consequences for Traffic Profiles, Measurement and Management in Broadband Access Platforms
4. Quality of Service Aspects for P2P and other Applications in Multi Service Networks