Dennis Edwards

Industrial tools for the feature location problem: an exploratory study

Software engineers who maintain and enhance large systems often encounter the feature location problem: where in the many thousands of lines of code is a particular user feature implemented? Several methods of addressing the problem have been proposed, most of which involve tracing the execution of the system and analyzing the traces. Some supporting academic tools are available. However, companies that depend on the successful evolution of large systems are more likely to use new methods if they are supported by industrial-strength tools of known reliability. This article describes a study performed with Motorola, Inc. to see whether there were any pitfalls in using Metrowerks CodeTEST and Klocwork inSight for feature location on message-passing software similar to systems that Motorola maintains. These two tools were combined with TraceGraph, an academic trace comparison tool. The study identified two main problems. First, some ‘glue’ code and workarounds were needed to get CodeTEST to generate a trace for an interval of time in which the feature was operating. Second, getting information out of TraceGraph and into inSight was needlessly complicated for a user. However, with a moderate amount of work, the tool combination was effective in locating, understanding and documenting features. Study participants completed these steps in typically 3–4 hours per feature, studying only a few hundred lines out of a 200 000 line system. An ongoing project with Motorola is focused on improving tool integration with the hope of making feature location common practice at Motorola. Copyright

Preventing unauthorized islanding: cyber-threat analysis

Intelligent software agents offer great potential in aiding the operation of electric power grids. These agents are networked applications that could be vulnerable to cyber attacks. Our goal is to design algorithms to detect or prevent such attacks. Before delving into this particular type of application, cyber attacks in general are filtered to develop a taxonomy of the types of threat that can occur. Any applications executing on a network have possible vulnerabilities that can be exploited by system hackers. Developing security solutions is an ongoing arms race between security professionals and hackers. Attack methods are generally unique to the application or system that are targeting, but underlying these strategies are common techniques. Although the purposes or consequences of attacks varies and are usually particular to the domain, the classes of consequences are limited. The national vulnerability database maintains reported vulnerabilities from industry, the private sector and government. By selecting a sampling of this vast database, we have formulated attack strategy categories and consequence categories. With these categories, a concise dissection of these specific type reports may be possible and an organized understanding of the vulnerabilities of networked applications results so threat prevention and detection algorithm can be developed

Software Feature Understanding in an Industrial Setting

Software Engineers frequently need to locate and understand the code that implements a specific user feature of a large system. This paper reports on a study by Motorola Inc. and the Software Engineering Research Center to see if industrial tools currently in use at the company could be adapted for feature understanding

Intelligent multi-agent framework for power system control and protection

In this paper, we propose a multi-agent methodology (MAM) for power system monitoring and protection. The uniqueness of the proposed architecture is the ability of MAM to evolve in the wake of an attack of malicious intent by mutation and thus always monitoring a power system bus remotely. Also, the architecture interacts with the mutated agents use a voting methodology thus alleviating the effect of agent malfunction in the presence of attack. This architecture also enables inter-agent communication which makes a collaborative framework and increases the degree of fault tolerance and leads to immediate problem solving. This architecture is tested on a two area network with five equivalent generators and 8 bus power networks for abnormal condition detection, generator bus isolation and also restoring power to the area which is affected by the bus isolation. Simulation studies indicate that the proposed methodology is capable of not only detecting the power grid attack but can also continuously monitor and isolate the generator bus even in the presence of sustained attack as the agents are regenerated.

Instrumenting time-sensitive software for feature location

Feature location is a commonly occurring program comprehension problem. In understanding large software systems, Software Engineers need to locate the different software components that work together to provide a specific end user feature.

Agents and islands: managing a power system before, during, and after transition to the islanded state

This paper presents a fictional scenario describing the effects of a category four hurricane on a metropolitan area, accompanied by a challenge: describe, and eventually realize, a system able to carry out the necessary power system operations without human participation. We outline the capabilities of an automated system for managing electric power. The overarching task of the automated system is islanding: To separate the metro areas power system from the primary power grid and manage its operation during several hurricane-induced contingencies, with the power system operational throughout. The essential technology needed to support this automation is agents. We address the roles agents play in the transition to the islanded state and in power system operation within the island; features of an appropriate agent substrate; the way the agents are organized; and information exchange among agents, the power system, and human operators

A Novel Self-Evolving Intelligent Multiagent Framework for Power System Control and Protection

In this paper, we propose a multiagent methodology (MAM) for power system monitoring and protection. The uniqueness of the proposed architecture is the ability of MAM to evolve in the wake of an attack of malicious intent by mutation, thus always monitoring power system buses remotely. The architecture interacts with the mutated agents using a voting methodology, thus alleviating the effect of agent malfunction in the presence of an attack. In addition, this proposed architecture enables inter-agent communication, thus developing a collaborative framework and increasing the degree of fault tolerance. This method is tested on a two-area five-machine eight-bus power network for abnormal condition detection and generator bus isolation, as well as for restoring power to the area that is affected by the bus isolation. Simulation studies indicate that the proposed methodology is not only capable of detecting the power grid attack but also capable of continuously monitoring and isolating the generator bus even in the presence of sustained attack as the agents are regenerated.

A novel self-evolving multi-agent architecture for power system monitoring and protection against attacks of malicious intent

In this paper, we propose a novel self-evolving multi-agent methodology (MAM) for power system monitoring and protection. The uniqueness of the proposed architecture is the ability of MAM to evolve in the wake of an attack of malicious intent by mutation and thus always monitoring a power system bus remotely. Also, the architecture interacting with the mutated agents uses a voting methodology thus alleviating the effect of agent malfunction in the presence of attack. This novel architecture is tested on a Single Machine Infinite Bus (SMIB) power network and a two area network with five equivalent generators and eight bus, two area power system for abnormal condition detection and generator bus isolation. Simulation studies indicate that the proposed methodology is capable of not only detecting the power grid attack once, but can also continuously monitor and isolate the generator bus even in the presence of sustained attack as the agents are regenerated.

Implementation and Validation of a Multi-Level Security Model Architecture

With the increasing complexity of transmissions grids and distribution networks, data communication security is becoming a paramount issue. This issue becomes more important in applications where software agent technology is being utilized, as software agents provide another possible channel of cyber attack. In this paper an innovative software agents-based multilevel security model architecture is presented. The goal of this architecture is to prevent known attacks, and to reduce or eliminate the consequences of successful attacks. An experimental protocol is presented to validate the developed security model architecture. Test results show the effectiveness of the developed architecture.

Communication analysis of distributed programs

Capturing and examining the causal and concurrent relationships of a distributed system is essential to a wide range of distributed systems applications. Many approaches to gathering this information rely on trace files of executions. The information obtained through tracing is limited to those executions observed. We present a methodology that analyzes the source code of the distributed system. Our analysis considers each processs source code and produces a single comprehensive graph of the systems possible behaviors. The graph, termed the partial order graph (POG), uniquely represents each possible partial order of the system. Causal and concurrent relationships can be extracted relative either to a particular partial order, which is synonymous to a single execution, or to a collection of partial orders. The graph provides a means of reasoning about the system in terms of relationships that will definitely occur, may possible occur, and will never occur. Distributed assert statements provide a means to monitor distributed system executions. By constructing the POG prior to system execution, the causality information provided by the POG enables run-time evaluation of the assert statement without relying on traces or addition messages.