SUBMITTED BY
Sanap Sushant L.

---

[attachment=11960]
ABSTRACT
Fault localization, a central aspect of network fault management, is a process of deducing the exact source of a failure from a set of observed failure indications. It has been a focus of research activity since the advent of modern communication systems, which produced numerous fault localization techniques. However, as communication systems evolved becoming more complex and offering new capabilities, the requirements imposed on fault localization techniques have changed as well.
Fault localization is a process of isolating faults responsible for the observable malfunctioning of the managed system. This paper reviews some existing approaches of this process and improves one of described techniques the probing. Probes are test transactions that can be actively selected and sent through the network. In this paper we use active probing to present an approach to develop tools for performing fault localization. We discuss various design issues involved and propose architecture for building such a tool.
Suggested innovations include: mixed (passive and active) probing, partitioning used for probe selection, logical detection of probing results, and adaptive, sequential probing.
CHAPTER 1: INTRODUCTION
1.1: What is Fault Localization?
Fault localization is a process of isolating faults responsible for the observable malfunctioning of the Managed system. Fault (also referred to as root problem) can be defined as an unpermitted deviation of at least one characteristic parameter or variable of a network object from acceptable or usual or standard values.
The International Standards Organization has divided network management tasks into six categories, as part of their Open System Interconnection Model. One of these categories the fault management can be characterized as detecting when network behavior deviates from normal and formulating a corrective course of action.
Fault management deals with
1. Fault detection, to know whether there is a failure or not in the network;
2. Fault localization, to know which is (are) the component(s) that has/have failed and caused the received alarms;
3. Fault isolation so that the network can continue to operate, which is the fast and automated way to restore Interrupted connections;
4. Network (re-)configuration that minimizes the impact of a fault by restoring the interrupted connections using spare equipment;
5. Replacement of the failing component(s).
Error, a consequence of fault, is defined as a discrepancy between observed and correct value. Fault may cause one or more errors.
Failure is an error that is visible to the outside world. Errors may propagate within the network causing failures of faultless hardware or software.
Symptoms are external manifestation of failures. They are observed (and send to the network manager) as alarms.
1.2: Need of Fault Localization
Since faults are unavoidable in communication systems, their quick detection and isolation is essential for the robustness, reliability, and accessibility of a system.
As computer networks increase in size, heterogeneity and complexity, effective management of such networks becomes more important and more difficult.
Network management is essential to ensure the good functioning of these networks. Fault diagnosis is a central aspect of network fault management. Since faults are unavoidable in communication systems, their quick detection and isolation is essential for the robustness, reliability, and accessibility of a system. In large and complex communication networks, automating fault diagnosis is critical.
Traditionally, fault localization has been performed manually by experts but, as systems grew larger and more complex, automated fault localization techniques became critical.
CHAPTER 2: FAULT LOCALIZATION TECHNIQUE
Introduction:
All techniques performing fault diagnosis rely on analysis of symptoms and events (such as warnings and parameters of the network elements) that are generated or detected during the occurrence of the fault. One can divide them in two main categories.
The first ones are Passive approaches, which compute fault location hypotheses on the basis of signals, generated by network elements by oneself and sent to management centers.
The second ones are Active approaches, which periodically check the state of the network elements, whether they are correct or not.
2.1assive fault localization techniques:
• Artificial intelligent (AI) methods
• Fault propagation methods
2.1.1: Artificial intelligent techniques for fault localization.
2.1.1.1: Model-based systems
Model-based systems construct an abstract model of the network. The model represents the network topology and is able to generate predictions of the normal behavior of the system. These predictions are compared with network observations and used for obtaining fault hypotheses. Depending of the kind of model, different approaches can be used: deterministic, probabilistic, temporal, finite state machines, etc. The advantages of these systems are that they are able to cope with incomplete information and with unforeseen failures. The drawback is the difficulty of developing good model for large networks and computation complexity.
2.1.1.2: Rule-based systems
Rule-based systems describe human expert knowledge in the form of decision rules, linking logical description of the network state (rules conditions) with partial or final localization hypotheses (rules conclusions).These systems do not require profound understanding of the architectural and operational principles of the network, and can effectively take human expertise into account. The disadvantages of rule-based systems are: the translation of human expertise into the set of rules, which cover all cases in an exhaustive manner is hard, and the need to search for all possible fault hypotheses slows down the global functioning of the system.
2.1.1.3: Case-based systems
Case-based systems make their decisions based on experience and past situations. They try to acquire relevant knowledge of past cases and previously used solutions to propose solutions for new problems. If these solutions cannot be taken directly from the case-base and need special reasoning on the base of closely matched situations, case based systems are computationally complex. Their advantages are efficiency and speed when the submitted problem was previously solved, and on-line learning that allows storing newly solved cases.
Some other AI techniques (neural networks, decision trees, etc.) are rarely used in these applications.
2.1.2: Fault propagation methods:
This family of techniques requires a priori specification of how a failure condition in one object is relevant to failure condition in other object.
2.1.2.1: Code-based techniques
Code-based techniques use causality graph model to describe the cause-and-effect relationships between network events. For each problem and each symptom a unique binary code is assigned, and fault propagation patterns are represented by a codebook. Fault localization is performed by finding a fault whose code is the closest match to the code of symptoms. For small systems this technique is very effective.
2.1.2.2: Dependency graph
Dependency graph is a directed graph whose nodes correspond to objects and whose edges denote the fact that a fault in starting object may cause a fault in ending object. Probabilities may be assigned to nodes and edges, describing uncertain relationships and events. Comparing a state of the graph with known state of the network one can find the source of fault symptoms.
2.2: Active fault localization techniques:
Active fault localization techniques construct managing tools which, instead of waiting for symptoms from the network, ask objects about their state and parameters. These techniques are not as popular as passive approaches but in some cases they may be very useful and therefore deserve attention.
2.2.1: Intelligent agents are simply software processes that live on every managed node, collecting, forwarding and setting management information, either at predefined intervals or when requested to by management station
2.2.2: Monitoring technique locates in some network nodes the computers (monitors) which are guaranteed by self-testing. Each monitor tests the adjacent nodes and links, and sends results of testing to the management station. Proper number of monitors can cover all nodes and links in the network. More advanced technique starts from only one monitor. Its adjacent nodes that pass the tests can became new monitors, then test their non-tested adjacent nodes and connected links, and so on.
2.2.3: Probing technique
Probing technique use an active measurement approach, called probing. Fault localization attempts to determine the state of the system from the probe results, so effectiveness of localization depends on the number of probes and their paths.