Root cause analysis includes the methods to identify the sources of errors in a network. Most techniques rely on knowledge models of the system, which are usually built by using network operators' expertise. This presents problems related to knowledge extraction, scalability, and understandability. We propose an offline method based on machine learning techniques for the automatic identification of dependencies between system events, enhanced with summarization, operations on graphs, and visualization that help network operators identify the root causes of errors. We illustrate it with examples from a corporate network.
Predicting failures on networks and systems is critical in order to maintain high uptime rates. Online failure prediction (OFP) techniques use machine learning and predictive analytics to generate failure models that can be applied to computer network data. These techniques can be provisioned on state-of-the-art stream processing systems, such as Spark Streaming, in order to cope with the scalability challenges from the base data. A big challenge with OFP is selecting the right information to process, as well as the appropriate features in order to achieve high accuracy in predicting failures on complex, interconnected systems. In this article we describe an OFP system built over Apache Spark that takes a repository of network management events, trains a Random Forest model, and uses this model to predict the appearance of future events in near real time. We show through our experiments the usefulness of network management events for accurate predictions, and the advantages of the proposed system in terms of predictive quality, cost, and ease of deployment.
1 January 2018 in Wireless Communications and Mobile Computing
In a more and more urbanized World, the so‐called Smart Cities need to be driven by the principles of efficiency and sustainability. Information and Communications Technologies and, in particular, the Internet of Things will play a key role on this, since they will allow monitoring and optimizing all the municipal services that exist and shall exist. People flow monitoring stands out in this context due to its wide range of applications, spanning from monitoring transport infrastructure to physical security applications. There are different techniques to perform people flow monitoring, presenting pros and cons, as in any other engineering problem. Typically, the options that provide the most accurate results are also the most expensive ones, whereas there are cases where presence detection in given areas is enough and cost is a limiting factor. The main goal of this paper is to prove that a minimal deployment of sensors, combined with the adequate analysis and visualization algorithms, can render useful results. In order to achieve this goal, a dataset is used with 1‐year data from a real infrastructure composed of 9 Wi‐Fi tracking sensors deployed in the Telecommunications Engineering School of Universidad Politécnica de Madrid, which is visited by 4000 people daily and covers 1.8 hectares. The data analysis includes time and occupancy, position of people, and identification of common behaviors, as well as a comparison of the accuracy of the considered solution with actual data and a video monitoring system available at the library of the school. The obtained insights can be used for optimizing the management and operation of the school, as well as for other similar infrastructures and, in general, for other kind of applications which require not very accurate people flow monitoring at low cost.
Failure prediction is a key component of modern autonomic systems. A crucial decision to take when performing it is which observation window to use, this is, to decide the time period in the past that will be taken into account in order to accurately predict. Currently, this decision is a highly manual process, dependent on expert knowledge. To alleviate this problem, we propose the usage of a customized genetic algorithm alongside a machine learning technique, random forests, which optimizes a novel, multiple observation window schemes that allows for more modeling complexity than other schemes present on the literature. We validate it using ten different events extracted from two real, industrial data sets: one from a high performance computing environment and one from a computer network. We show that our algorithm creates models that optimize performance while reducing the observed time automatically with minimal user input required.
Infrastructure usage discovery, positioning and analysis of behaviors of its users, usually requires a collection of accurate and frequent positioning data. This paper shows how a network of inexpensive and non-intrusive sensors can serve to perform this kind of analysis by detecting devices with wifi connectivity. We are able to obtain the hours of use of the infrastructures, the occupation of the different areas at each moment and some of the most common users' behaviours.