
Microservices architecture is becoming popular because it is flexible and can easily grow with demand. But sometimes, resources are not used properly because of poor task scheduling. This research suggests a better task scheduling method to make resource use more efficient, reduce delays, and improve overall system performance. The proposed method uses dynamic task prioritization and load balancing to make sure resources are used in the best way. The results show a clear improvement in CPU and memory usage without affecting system performance. This approach can be used in many types of microservice systems to make them faster and more efficient. It also works well in environments where different systems have different capacities and performances. The scheduling algorithm introduced here focuses on working together to manage all resources effectively. It looks at both CPU use and network traffic, which helps achieve better resource allocation, improved service performance, and a more cost-effective and responsive microservices system.
The rapid proliferation of Renewable Energy Sources (RES), particularly Solar Photovoltaic (SPV) systems, into modern transmission networks has intensified concerns regarding voltage stability, power loss management, and grid reliability under increasing load demand. This paper presents a systematic technical framework for the optimal siting and sizing of SPV systems in large-scale transmission networks using the Voltage Stability Index (VSI) as a bus vulnerability indicator. Candidate buses prone to voltage collapse are identified through VSI-based ranking, and subsequently, an optimal SPV allocation strategy is formulated to maximise voltage profile improvement and minimise real power losses. Simulations are conducted in ETAP 19.1 employing the adaptive Newton-Raphson load flow algorithm on the standard IEEE 57-bus test system, encompassing ten case studies—four single SPV plant configurations and six dual SPV plant configurations. The proposed methodology consistently yielded significant enhancement in nodal voltage profiles and system-wide loss reduction across all tested scenarios. The maximum active power loss reduction achieved was 34.68
Software-aging studies in mobile systems often focus on internal resource trends, even though an operational effect may be a user-perceived decline in responsiveness. This study presents SAFR (Software Aging Forensic Reconstruction), a logcat-driven framework that uses post-rejuvenation application launch latency as its principal response. Rather than assuming that all launch records are comparable, SAFR reconstructs the preceding package force-stop, process creation, launcher request, displayed activity, and available runtime evidence before temporal analysis. SAFR was evaluated on a Google Pixel 5 running Android 14 under a sustained Android Monkey workload involving 20 applications and periodic application rejuvenation. The preserved trace contained 22,128,594 logcat lines spanning 41.016 h, including 18.427 h of active rejuvenation. The framework reconstructed 1072 complete rejuvenation batches, 11,590 raw Android Displayed records, and 10,914 eligible post-rejuvenation launch chains. After autocorrelation screening, the Hamed–Rao modified Mann–Kendall procedure showed no significant monotonic trend in the unadjusted cycle-level series. In the context-adjusted model, cycle progression remained positively associated with launch latency after adjustment for skin temperature, garbage-collection count and duration, and Monkey workload intensity. The estimated coefficient was 0.1007 ms per cycle (p = 6.54 × 10⁻⁸; 95
Brain tumors, arising from the abnormal proliferation of cells within brain tissue, represent one of the most critical and life-threatening medical conditions. Early detection plays a pivotal role in improving clinical outcomes and increasing survival rates. Although radiologists can identify tumors through manual inspection of Magnetic Resonance Imaging (MRI) scans, this process is often time-consuming, labor-intensive, and prone to human error, particularly when dealing with large datasets. Consequently, deep learning–based automated diagnostic tools have emerged as promising alternatives to enhance both accuracy and efficiency. In this study, we propose a transfer learning–based approach for the automatic classification of brain tumors. The methodology incorporates several pre-trained Convolutional Neural Network (CNN) architectures, including VGG16, VGG19, InceptionResNetV2, Xception, DenseNet121, DenseNet169, DenseNet201, MobileNet, and MobileNetV2, which are fine-tuned to distinguish between two categories: Negative (No tumor) and Positive (Tumor). To improve transparency and foster trust in model predictions, we employed two eXplainable Artificial Intelligence (XAI) techniques: Gradient-weighted Class Activation Mapping (Grad-CAM) and Local Interpretable Model-agnostic Explanations (LIME). These methods highlight the critical regions that influence classification decisions. Experimental results demonstrate that the DenseNet121 model achieved the best overall performance, with an accuracy of 94.91