Logo des Repositoriums

Auflistung nach Schlagwort "prediction"

1 - 10 von 13
  • Konferenzbeitrag
    Commit-Based Continuous Integration of Performance Models
    (Softwaretechnik-Trends Band 43, Heft 2, 2023) Armbruster, Martin
    Architecture-level performance models (aPM) such as the Palladio Component Model (PCM) can be used for, e.g., performance predictions to explore design alternatives and combines the aspects of architecture and performance models. An up-to-date architecture model can support the software maintenance by reducing the architectural degradation or guide the software evolution. At the same time, performance models allow the investigation of the software performance without the need to implement or change the system. However, keeping them up-to-date requires manual effort which hinders their adoption. Especially in the agile software development which is characterized by incremental and iterative development cycles, no or short design phases prevent manual modeling activities. This master thesis presents an approach building upon the previous work with these two main goals: (1) closing the gaps by completing the pipeline for the aPM extraction and instrumentation, and (2) evaluating the pipeline with a real Git history. As a result, in the approach, the Java source code in the state of a new commit is parsed into a code model. By a state-based comparison with the code model of the previous commit, a delta-based change sequence is obtained which describes how the code model of the previous commit can be transformed into a code model conforming to the state of the new commit. Afterwards, the changes are utilized to incrementally update the PCM, i.e., only the PCM elements affected by the source code changes are updated while the other PCM elements remain unchanged.
  • Workshopbeitrag
    Effective Toxicity Prediction in Online Multiplayer Gaming: Four Obstacles to Making Approaches Usable
    (Mensch und Computer 2022 - Workshopband, 2022) Frommel, Julian; Mandryk, Regan
    Toxicity represents a threat to the safety and health of online multiplayer gaming communities. This has been recognized by industry, academia, and players and led to efforts for combating toxicity, including different approaches for predicting toxicity from behaviour. Despite promising results, such approaches have not yet been able to meaningfully combat toxicity at scale. In this position paper, we describe four obstacles that impede usable applied toxicity prediction in multiplayer games that could help to combat harm.We want to foster a discussion about how user-centered artificial intelligence approaches may help solve these obstacles.
  • Konferenzbeitrag
    Mapping Data Flow Models to the Palladio Component Model
    (Softwaretechnik-Trends Band 39, Heft 4, 2019) Seifermann, Stephan; Werle, Dominik; Ebada, Mazen
    Predicting quality properties such as privacy are reasonable use cases for Data Flow Models (DFMs). For other use cases such as performance prediction, component-based software architecture models focusing on control flows are more suitable. Designers can derive a Control Flow Model (CFM) from a DFM but they have to make numerous design decisions like defining operation signatures. Currently, this derivation is a creative process without a clear design space and without guidelines for navigating this space. In this paper, we present design alternatives for given data flow examples and derive mapping rules that allow to choose between reasonable alternatives. Our results are a first step towards a catalogue of rules for deriving CFMs from DFMs in a systematic way and providing semi-automated transformations.
  • Konferenzbeitrag
    Modelling and Predicting Memory Behaviour in Parallel Systems with Network Links? Palladio-based Experiment Report
    (Softwaretechnik-Trends Band 39, Heft 4, 2019) Gruber, Philipp; Frank, Markus
    This work improves the capabilities of Palladio to predict the performance of parallel software in multicore environments. In previous work, we could show that the accuracy of the Palladio simulations is not sufficient for multicore systems. We assume that one reason for this is the memory bandwidth behaviour, which is not included in the Palladio Component Model and can become a bottleneck in parallel software. We present an approach to model the memory bandwidth behaviour by the means of an already existing network link concept. We can show that by using network link as a memory model we can improve our predictions up to 26% points using 16 cores on a machine and can receive an accuracy of 90% for our use case.
  • Konferenzbeitrag
    Multi-level Hardware Simulation in Palladio
    (Softwaretechnik-Trends Band 43, Heft 4, 2023) Weber, Sebastian; Taghavi, Bahareh
    Predicting the fulfillment of quality requirements, e.g., performance, is important during design and implementation of a software system to ensure its implementation can meet these requirements. Model-based analysis is a common approach to get such predictions. It usually requires a trade-off between the accuracy of the result and the execution time of the analysis, because resources for the execution of the analysis are limited, but some components might require a more accurate analysis. When looking at traditional model-based analyses, like the Palladio approach, the user can decide this trade-off and with it the level of granularity of the model and simulation only prior to, but not throughout, the simulation execution. In addition, the complexity and heterogeneity of systems is increasing which complicates modelling and analysis on a single level further. This paper presents a concept for the extension of the Palladio approach to support modelling and analysis of systems on multiple levels of abstraction to overcome these limitations.
  • Konferenzbeitrag
    Performance Modelling of Message-Oriented Middleware with Priority Queues
    (Softwaretechnik-Trends Band 40, Heft 3, 2020) Singh, Snigdha; Schmid, Larissa; Koziolek, Anne
    Message-Oriented Middleware (MOM) with priority queues reduces the latency of critical events. In general, MOM uses a FIFO queuing methodology. But, different application scenarios require certain critical events with higher priority to be served earlier over low-priority events, so that the subscriber of the event consumes the high-priority event with less delay. In the context of the Palladio Component Model (PCM), MOM-based systems have been modelled considering message queue length and latency as metrics for performance prediction and simulation. However, the approaches did not consider modelling MOM with priority queues and their impact on performance. We will first, discuss the existing approaches in PCM which support performance prediction for MOM-based systems and then propose how they can be extended to support performance predictions for MOM with priority queuing. We will then conclude which approach is best suited to extend by assessing their capabilities to predict performance metrics relevant for priority queuing, especially the delay of individual events at the subscriber end.
  • Konferenzbeitrag
    Predicting Scaling Efficiency of Distributed Stream Processing Systems via Task Level Performance Simulation
    (Softwaretechnik-Trends Band 43, Heft 1, 2023) Rank, Johannes; Barnert, Maximilian; Hein, Andreas; Krcmar, Helmut
    Stream processing systems (SPS) are a special class of Big Data systems that firms employ in (near) real time business scenarios. They ensure low-latency processing through a high degree of parallelization and elasticity. However, firms often do not know which scaling direction: horizontally, vertically, or mixed, is the best strategy in terms of CPU performance to scale those systems. Especially in cloud deployments with a pay-per-use model and cluster sizes that can span dozens of cores and machines, firms would profit from more accurate measurement-based approaches. In this paper, we show how to predict the CPU consumption of Apache Flink for different scaling scenarios using the Palladio Component Model. Our approach models the individual streaming tasks that make up the application and parametrizes it with fine grained CPU metrics obtained by combining BPF pro filing and querying the CPU’s performance measurement unit. Through this “task-level model approach”, we can achieve highly accurate predictions, despite using a simple model and only requiring a few mea surements for parametrization. Our experiment also shows that we achieve more accurate results than an alternative approach based on regression analysis.
  • Konferenzbeitrag
    Recovering Missing Dependencies in Java Models
    (Softwaretechnik-Trends Band 43, Heft 4, 2023) Armbruster, Martin; Mazkatli, Manar; Koziolek, Anne
    Different approaches use models of source code to extract performance models from the code which allow performance predictions and the exploration of design alternatives. The extended Java Model Parser and Printer provides a modeling environment for Java code. It defines a metamodel and contains a parser and printer including three variants to resolve references between different Java models. These variants assume that the complete code with all dependencies is available or missing elements are not accessed. In this paper, a trivial recovery strategy is introduced. It is able to recover references which cannot be resolved. Additionally, the performance and model storage of the reference resolution’s variants are compared with and without the trivial recovery. The results indicate that the trivial recovery reduces the execution time and required space for storing the models. In the future, further recovery strategies can be implemented to allow a balance between performance and model accuracy.
  • Konferenzbeitrag
    The relation of convergent thinking and trace data in an online course
    (DELFI 2021, 2021) Rüdian, Sylvio; Haase, Jennifer; Pinkwart, Niels
    Many prediction tasks can be done based on users’ trace data. In this paper, we explored convergent thinking as a personality-related attribute and its relation to features gathered in interactive and non-interactive tasks of an online course. This is an under-utilized attribute that could be used for adapting online courses according to the creativity level to enhance the motivation of learners. Therefore, we used the logfile data of a 60 minutes Moodle course with N=128 learners, combined with the Remote Associates Test (RAT). We explored the trace data and found a weak correlation between interactive tasks and the RAT score, which was the highest considering the overall dataset. We trained a Random Forest Regressor to predict convergent thinking based on the trace data and analyzed the feature importance. The result has shown that the interactive tasks have the highest importance in prediction, but the accuracy is very low. We discuss the potential for personalizing online courses and address further steps to improve the applicability.
  • Konferenzbeitrag
    Supporting Backward Transitions within Markov Chains when Modeling Complex User Behavior in the Palladio Component Model
    (Softwaretechnik-Trends Band 40, Heft 3, 2020) Barnert, Maximilian; Krcmar, Helmut
    The specification of complex user behavior as accurate as possible is required in order to evaluate performance characteristics for application systems. Approaches exist to model probabilistic aspects within user behavior for session-based application systems using Markov chains. To integrate these approach into performance prediction activities, the authors transform the workload specifications of WESSBAS into performance model instances of the Palladio Component Model (PCM). This paper presents our approach to enable backward transitions within Markov chains using available elements of the PCM meta-model. By extending the existing approach, further complexity within workload for application systems is supported during performance modeling.