Dustminer

2012

A key challenge in developing large scale applications (both in system size and in input size) is finding bugs that are latent at the small scales of testing, only manifesting when a program is deployed at large scales. Traditional statistical techniques fail because no error-free run is available at deployment scales for training purposes. Prior work used scaling models to detect anomalous behavior at large scales without being trained on correct behavior at that scale. However, that work cannot localize bugs automatically. In this paper, we extend that work with automatic diagnosis technique, based on feature reconstruction, and validate our design through case studies with two real bugs from an MPI library and a DHT-based file sharing application.

REV Journal on Electronics and Communications, 2016

Fault analysis in communication networks and distributed systems is a difficult process that heavily depends on system administrator’s experience and supporting tools. This process usually requires analytic techniques and several types of event data including log events, debug messages, trace obtained from these systems to investigate the root cause of faults. This paper introduces an approach of exploiting context-aware data and classification technique for improving this process. This approach uses both event data and context-aware data including CPU load, memory, processes, temperature, status to train a decision tree, and then applies the tree to assess suspected events. We have implemented and experimented the approach on the OpenStack cloud computing system with the Hadoop computing service and MELA event collection system. The experimental results reveal that the accuracy score of the approach reaches 85% on average. The paper also includes detailed analysis for the results.

Proceedings of the tenth ACM international conference on Embedded software - EMSOFT '12, 2012

Increasing complexity in both the software and the underlying hardware, and ever tighter time-to-market pressures are some of the key challenges faced when designing multimedia embedded systems. Optimizing the debugging phase can help to reduce development time significantly. A powerful approach used extensively during this phase is the analysis of execution traces. However, huge trace volumes make manual trace analysis unmanageable. In such situations, Data Mining can help by automatically discovering interesting patterns in large amounts of data. In this paper, we are interested in discovering periodic behaviors in multimedia applications. Therefore, we propose a new pattern mining approach for automatically discovering all periodic patterns occurring in a multimedia application execution trace. Furthermore, gaps in the periodicity are of special interest since they can correspond to cracks or drop-outs in the stream. Existing periodic pattern definitions are too restrictive regarding the size of the gaps in the periodicity. So, in this paper, we specify a new definition of frequent periodic patterns that removes this limitation. Moreover, in order to simplify the analysis of the set of frequent periodic patterns we propose two complementary approaches: (a) a lossless representation that reduces the size of the set and facilitates its analysis, and (b) a tool to identify pairs of "competitors" where a pattern breaks the periodicity of another pattern. Several experiments were carried out on embedded video and audio decoding application traces, demonstrating that using these new patterns it is possible to identify abnormal behaviors.

2006

Recent experience with the deployment of sensor networks demonstrates that it is far from trivial to setup a working larger-scale sensor network in the field. Even though simulations and experiments with lab testbeds confirmed a working system, subtle real-world influences lead to frequent failures in the field. Identifying and fixing these problems in the field is currently a difficult and cumbersome task due to the lack of appropriate concepts and tools. In this paper we address this issue by, firstly, classifying common problems that have been encountered during deployment. We then show that many of these problems can be detected by overhearing and analyzing sensor network traffic without need for an instrumentation of sensor nodes. Based on this observation, we develop a tool to inspect a deployed sensor network, consisting of a distributed network sniffer and a data-stream-based framework for online traffic analysis. We demonstrate and evaluate how this tool can be used to debug a typical data gathering application. 1

This paper proposes an interactive debugging aid that exhibits a deep understanding of a narrow class of bugs. This system, called Sniffer, will be able to find and identify errors, and explain them in terms which are relevant to the programmer. Sniffer is knowledgeable about side-effects. It is capable of citing the data which was in effect at the time an error became manifest. The debugging knowledge in Sniffer is organized as a collection of independent experts which know about particular errors. The experts (sniffers) perform their function by applying a feature recognition process to the text for the program, and to the events which took place during the execution of the code. No deductive machinery is involved. The experts are supported by two systems; the cliche finder which identifies small portions of algorithms from a plan for the code, and the time rover which provides complete access to all program states that ever existed. Sniffer is embedded in a run-time debugging aid...

2005

Abstract required to diagnose the failure, ie, to identify the source of the failure. Diagnosis is challenging because fast error propagation may occur in high throughput distributed applications. The diagnosis often needs to be probabilistic in nature due to imperfect observability of the payload system, inability to do white-box testing, constraints on the amount of state that can be maintained at the diagnostic process, and imperfect tests used to verify the system.

2002

Developers and users of high-performance distributed systems often observe performance problems such as unexpectedly low throughput or high latency. Determining the source of the performance problems requires detailed end-to-end instrumentation of all components, including the applications, operating systems, hosts, and networks. In this paper we describe a methodology that enables the real-time diagnosis of performance problems in complex high-performance distributed systems. The methodology includes tools for generating timestamped event logs that can be used to provide detailed end-to-end application and system level monitoring; and tools for visualizing the log data and real-time state of the distributed system. This methodology, called NetLogger, has proven invaluable for diagnosing problems in networks and in distributed systems code. This approach is novel in that it combines network, host, and application-level monitoring, providing a complete view of the entire system. NetLogger is designed to be extremely light-weight, and includes a mechanism for reliably collecting monitoring events from multiple distributed locations. This technical report summarizes most important points of several previous papers on NetLogger, and is meant to be used as a general overview.

2008

Network diagnosis, an essential research topic for traditional networking systems, has not received much attention for wireless sensor networks. Existing sensor debugging tools like sympathy or EmStar rely heavily on an add-in protocol that generates and reports a large amount of status information from individual sensor nodes, introducing network overhead to a resource constrained and usually traffic sensitive sensor network. We report in this study our initial attempt at providing a light-weight network diagnosis mechanism for sensor networks. We propose PAD, a probabilistic diagnosis approach for inferring the root causes of abnormal phenomena. PAD employs a packet marking algorithm for efficiently constructing and dynamically maintaining the inference model. Our approach does not incur additional traffic overhead for collecting desired information. Instead, we introduce a probabilistic inference model which encodes internal dependencies among different network elements, for online diagnosis of an operational sensor network system. Such a model is capable of additively reasoning root causes based on passively observed symptoms. We implement the PAD design in our sea monitoring sensor network test-bed and validate its effectiveness. We further evaluate the efficiency and scalability of this design through extensive trace-driven simulations.

Lecture Notes in Computer Science, 2010

While discrete event systems have been widely applied for diagnosing distributed communicating systems, existing models may not completely satisfy the requirements for the application of fault identification and repair in software systems. This paper presents a model-based diagnosis approach that identifies possible faults based on generic fault models in abstract traces where events may be associated to multiple system components. We overcome the common limitation that precise fault models are available for each component and leverage generic fault models of classes of faults instead. We show that diagnoses representing entire classes of equivalent solutions can be computed based on local information and investigate the performance of our algorithm.

2016

Software diagnosis is a particularly challenging problem for modern systems, which may consist of dozens, if not hundreds, of components computing on concurrent and potentially distributed platforms, and using infrastructure and services built by many organizations. We propose a framework that generalizes state-of-the-art classical reasoning-based fault diagnosis which tolerates observation uncertainty and addresses degradation of quality of service. Empirical evaluation involving 27 000 highly realistic synthetic scenarios demonstrates an average accuracy improvement of 20% (with 99% statistical significance) which is considerable in the domain of Software Fault Localization (SFL). We measure the improvement in accuracy on well-established SFL performance metrics. Introduction One of the most important way to improve the trustworthiness of software systems is to increase their robustness in the face of (runtime) failures. While design-time methods are useful in improving confidence...