To date, several explainable AI methods are available. The variability of the resulting explanations can be high, especially when many input features are considered. This lack of robustness may limit their usability. In this paper we try to fill this gap, by contributing a methodology that: i) is able to measure the robustness of a given set of explanations; ii) suggests how to improve robustness, by tuning the model parameters. Without loss of generality, we exemplify our proposal for ensemble tree models, which typically reach a high predictive performance in classification problems. We consider a toy case study with artificially generated data as well as two real case studies whose application domain is cybersecurity and more precisely the models used for detecting phishing websites.

Modern IoT ecosystems face many security issues. An aspect often neglected concerns covert channels, which allow for exfiltrating data or preventing detection. To this aim, the Message Queuing Telemetry Transport (MQTT) protocol can be abused to create various hidden communication paths, mainly due to its textual nature. Alas, simpler detection metrics could be ineffective and their optimization requires a vast number of test cases. Therefore, this paper proposes to use a small language model trained over real MQTT topics to automatically generate the required test cases. Results indicate the need for optimizations to make popular detection metrics usable “in the wild”.

Abstract Virtualization and server consolidation are the technologies that govern today’s data centers, allowing both efficient management at the functionality level as well as at the energy and performance levels. There are two main ways to virtualize either using virtual machines or containers. Both have a series of characteristics and applications, sometimes being not compatible with each other. Not to lose the advantages of each of them, there is a trend to load data centers by nesting containers in virtual machines. Although there are good experiences at a functional level, the performance and energy consumption trade-off of these solutions is not completely clear. Therefore, it is necessary to study how this new trend affects both energy consumption and performance. In this work, we present an experimental study aimed to investigate the behavior of nesting containers in virtual machines while executing CPU-intensive workloads. Our objective is to understand what performance and energy nesting configurations are equivalent or not. In this way, administrators will be able to manage their data centers more efficiently.

Summary Cloud computing provides cost‐effective solutions for deploying services and applications. Although resources can be provisioned on demand, they need to adapt quickly and in a seamless way to the workload intensity and characteristics and satisfy at the same time the desired performance levels. In this paper, we evaluate the effects exercised by different incoming workload patterns on cloud autoscaling strategies. More specifically, we focus on workloads characterized by periodic, continuously growing, diurnal and unpredictable arrival patterns. To test these workloads, we simulate a realistic cloud infrastructure using customized extensions of the CloudSim simulation toolkit. The simulation experiments allow us to evaluate the cloud performance under different workload conditions and assess the benefits of autoscaling policies as well as the effects of their configuration settings.

The services and applications deployed nowadays in cloud environments are characterized by variable intensity and resource requirements. The variability of these workloads coupled with their heterogeneity affects the cost associated with the cloud infrastructure and the performance levels that can be satisfied. In these complex scenarios, resource provisioning policies have to take into account the actual workloads being processed and pro-actively anticipate in a timely manner the changes in workload intensity and characteristics. To support this decision process, we propose an integrated approach - that combines various workload characterization techniques - for modeling and predicting workload access patterns. The application of this approach has shown the importance of identifying models that specifically capture and reproduce the dynamics of these patterns and consider at the same time their peculiarities.

Cloud computing allows users to devise cost-effective solutions for deploying their applications. Nevertheless, the decisions about resource provisioning are very challenging because workloads are seriously affected by the uncertainty of cloud performance and their characteristics vary. In this paper we address these issues by explicitly modeling workload and cloud uncertainty in the decision process. For this purpose, we adopt a probabilistic formulation of the optimization problem aimed at minimizing the expected cost for deploying a parallel application under a deadline constraint. To find a sub-optimal solution of the problem we apply a Genetic Algorithm. By tuning its parameters we are able to assess their role and their impact on the effectiveness and efficiency of the algorithm for provisioning and scheduling in uncertain cloud environments.

Online social systems have emerged as a popular medium for people in society to communicate with each other. Among the most important reasons why people communicate is to share emotional problems, but most online social systems are uncomfortable or unsafe spaces for this purpose. This has led to the rise of online emotional support systems, where users needing to speak to someone can anonymously connect to a crowd of trained listeners for a one-on-one conversation. To better understand who, how, and when users utilize emotional support systems, this paper examines user and conversation characteristics on 7 Cups of Tea. 7 Cups of Tea is a massive, vibrant emotional support system with a community of listeners ready to help those with any number of emotional issues. Intriguing insights, including evidence of world-wide adoption of the service, the need to seek immediate support from many others, and a rich-get-richer phenomenon underscore a growing need for online emotional support systems and highlight important aspects to promote their long-term viability.

Resource provisioning and task scheduling in Cloud environments are quite challenging because of the fluctuating workload patterns and of the unpredictable behaviors and unstable performance of the infrastructure. It is therefore important to properly master the uncertainties associated with Cloud workloads and infrastructure. In this paper, we propose a probabilistic approach for resource provisioning and task scheduling that allows users to estimate in advance, i.e., offline, the resources to be provisioned, thus reducing the risk and the impact of overprovisioning or underprovisioning. In particular, we formulate an optimization problem whose objective is to identify scheduling plans that minimize the overall monetary cost for leasing Cloud resources subject to some workload constraints. This cost-aware model ensures that the execution time of an application does not exceed with a given probability a specified deadline, even in presence of uncertainties. To evaluate the behavior and sensitivity to uncertainties of the proposed approach, we simulate a simple batch workload consisting of MapReduce jobs. The experimental results show that, despite the provisioning and scheduling approaches that do not take into account the uncertainties in their decision process, our probabilistic approach nicely adapts to workload and Cloud uncertainties.

Internet and online-based social systems are rising as the dominant mode of communication in society. However, the public or semi-private environment under which most online communications operate under do not make them suitable channels for speaking with others about personal or emotional problems. This has led to the emergence of online platforms for emotional support offering free, anonymous, and confidential conversations with live listeners. Yet very little is known about the way these platforms are utilized, and if their features and design foster strong user engagement. This paper explores the utilization and the interaction features of hundreds of thousands of users on 7 Cups of Tea, a leading online platform offering online emotional support. It dissects the user's activity levels, the patterns by which they engage in conversation with each other, and uses machine learning methods to find factors promoting engagement. The study may be the first to measure activities and interactions in a large-scale online social system that fosters peer-to-peer emotional support.

The technologies aimed at Web content discovery, retrieval and management face the compelling need of coping with its highly dynamic nature coupled with complex user interactions. This paper analyzes the temporal patterns of the content changes of three major news websites with the objective of modeling and predicting their dynamics. It has been observed that changes are characterized by a time dependent behavior with large fluctuations and significant differences across hours and days. To explain this behavior, we represent the change patterns as time series. The trend and seasonal components of the observed time series capture the weekly and daily periodicity, whereas the irregular components take into account the remaining fluctuations. Models based on trigonometric polynomials and ARMA components accurately reproduce the dynamics of the empirical change patterns and provide extrapolations into the future to be used for forecasting.

News websites are expected to deliver in a timely manner the latest stories as well as their latest developments. Thereby, tools, such as, search engines, need to cope with these rapid and frequent content changes by adjusting their crawling activities accordingly. In this paper we explore and model the properties and temporal behavior of the content changes of three major news websites. The dynamics of the changes is characterized by large fluctuations and significant differences from day to day and from hour to hour. However, a certain degree of similarity in the overall patterns of each website exists. In particular, the application of multivariate analysis techniques allows us to identify groups of days with similar change patterns, thus allowing for the customization of the crawling policies adopted by search engines.

The traffic produced by the periodic crawling activities of Web robots often represents a good fraction of the overall websites traffic, thus causing some non-negligible effects on their performance. Our study focuses on the traffic generated on the SPEC website by many different Web robots, including, among the others, the robots employed by some popular search engines. This extensive investigation shows that the behavior and crawling patterns of the robots vary significantly in terms of requests, resources and clients involved in their crawling activities. Some robots tend to concentrate their requests in short periods of time and follow some sorts of deterministic patterns characterized by multiple peaks. The requests of other robots exhibit a time dependent behavior and repeated patterns with some periodicity. We represent the traffic as a time series modelled in the frequency domain. The identified models, consisting of trigonometric polynomials and Auto Regressive Moving Average components, accurately summarize the behavior of the overall traffic as well as the traffic of individual robots. These models can be easily used as a basis for forecasting.

The content of news websites changes frequently and rapidly and its relevance tends to decay with time. To be of any value to the users, tools, such as, search engines, have to cope with the dynamics of websites and detect changes in a timely manner. In this paper we apply time series analysis to study the properties and the temporal patterns of the change rates of the content of three news websites. Our investigation shows that changes are characterized by large fluctuations with periodic patterns and time dependent behavior. The time series describing the change rate is decomposed into trend, seasonal and irregular components and models of each component are then identified. The trend and seasonal components describe the daily and weekly patterns of the change rates. Trigonometric polynomials best fit these deterministic components, whereas the class of ARMA models represents the irregular component. The resulting models can be used to describe the dynamics of websites and predict future change rates.

High-performance computing (HPC) programmers utilize tracefiles, which record program behavior in great detail, as the basis for many performance analysis activities. The lack of generally accessible tracefiles has forced programmers to develop their own testbeds in order to study the basic performance characteristics of the platforms they use. Since tracefiles serve as input to performance analysis and performance prediction tools, tool developers have also been hindered by the lack of a testbed for verifying and fine-tuning tool functionality, We created a community repository that meets the needs of both application and tool developers. In this paper, we describe how the tracefile testbed was designed to facilitate flexible searching and retrieval of tracefiles based on a variety of characteristics. Its Web-based interface provides a convenient mechanism for browsing, downloading, and uploading collections of tracefiles and tracefile segments, as well as viewing statistical summaries of performance characteristics.

The performance of a system is determined by its characteristics as well as by the composition of the load being processed. Hence, its quantitative description is a fundamental part of all performance evaluation studies. Several methodologies for the construction of workload models, which are functions of the objective of the study, of the architecture of the system to be analyzed, and of the techniques adopted, are presented. A survey of a few applications of these methodologies to various types of systems (i.e., batch, interactive, database, network-based, parallel, supercomputer), is given.>

In a paper published in 1984 [Ferr84], the validity of applying clustering techniques to the design of an executable model for an interactive workload was discussed. The following assumptions, intended not to be necessarily realistic but to provide sufficient conditions for the applicability of clustering techniques, were made:

The knowledge of workload fluctuations is fundamental in all performance studies in which the dynantic characteristics of the resource demands must be taken into account. Among the several workload data sequences that may be considered, the arrival pattern of workload components is certainly one of the most important. An approach to the identification of the arrival rate functions through a numerical fitting technique that allows one to have concise representations of the arrival patterns during one-day periods is presented. The variability of the arrival pattern over different days is also investigated. The patterns which may be considered as "representatives" of the analyzed workload are found through the application of the clustering technique. The parametric model of the arrival process may easily be used to forecast the load of the system in the near future or to drive system simulations when several dynamic control policies are to be investigated. The proposed modeling approach has been validated, in the sense of establishing its predictive value, through the analysis of workload data for the same installation collected during different periods.

Applications of modeling techniques based on queueing theory to computer system performance analysis normally assume the existence of steady-state conditions. However, these conditions are often violated since the unpredictable composition of workload causes peaks having highly variable intensities and durations. Furthermore, computer system performance is highly dependent on how the system reacts to workload fluctuations. Automatic control mechanisms are required to take care of the high variance of resource demands. Real-time optimization of the overall performance of a computer system requires the introduction of adaptive control on the controlled functions, An adaptive scheduling algorithm which controls the input of the system in order to maximize a given performance criterion, such as the system throughput, is presented. The system load is adjusted depending on the characteristics of both the mix of jobs in execution and the mix of jobs submitted to the system and waiting in the input queue. The asymptotic analysis of the performance bounds provides useful information about the limits on the performance indexes that can be achieved with a multiclass workload. The evaluation of the adaptive control system is performed through simulation experiments using data collected from two real workloads. This technique could be used to optimize the throughput of a centralized system as well as for the automatic load balancing in a distributed environment.