arXiv subjects

Duneesha Fernando

Publications and source records attributed to Duneesha Fernando.

4 recordsLinked to original sources

arXiv2026

A Cascaded Graph Neural Network for Joint Root Cause Localization and Analysis in Edge Computing Environments

Edge computing environments host increasingly complex microservice-based IoT applications that are prone to performance anomalies propagating across dependent services. Identifying the faulty component (root cause localization) and the underlying fault type (root cause analysis) is essential for timely mitigation. Supervised graph neural networks (GNNs) currently represent the state of the art for joint root cause localization and analysis. However, existing approaches rely on centralized processing over full-system graphs, leading to high inference latency and limited scalability in large, distributed edge environments. In this paper, we propose a cascaded GNN framework for joint RCL and fault type identification that explicitly addresses these scalability challenges. Our approach employs communication-driven clustering to partition large service graphs into highly interacting communities and a cascaded network with two subnetworks that perform hierarchical RCL/RCA. By restricting message passing to reduced and structured subgraphs, the proposed framework significantly lowers computational complexity while preserving critical dependency information. We evaluate the proposed method on the MicroCERCL benchmark and large-scale datasets generated using the iAnomaly simulation framework. Experimental results show that the cascaded architecture achieves diagnostic accuracy comparable to centralized GNN baselines while maintaining near-constant inference latency as graph size increases, enabling scalable and actionable AIOps in edge computing environments.

cs.DC↗

arXiv2025

A Decentralized Root Cause Localization Approach for Edge Computing Environments

Edge computing environments host increasingly complex microservice-based IoT applications, which are prone to performance anomalies that can propagate across dependent services. Identifying the true source of such anomalies, known as Root Cause Localization (RCL), is essential for timely mitigation. However, existing RCL approaches are designed for cloud environments and rely on centralized analysis, which increases latency and communication overhead when applied at the edge. This paper proposes a decentralized RCL approach that executes localization directly at the edge device level using the Personalized PageRank (PPR) algorithm. The proposed method first groups microservices into communication- and colocation-aware clusters, thereby confining most anomaly propagation within cluster boundaries. Within each cluster, PPR is executed locally to identify the root cause, significantly reducing localization time. For the rare cases where anomalies propagate across clusters, we introduce an inter-cluster peer-to-peer approximation process, enabling lightweight coordination among clusters with minimal communication overhead. To enhance the accuracy of localization in heterogeneous edge environments, we also propose a novel anomaly scoring mechanism tailored to the diverse anomaly triggers that arise across microservice, device, and network layers. Evaluation results on the publicly available edge dataset, MicroCERCL, demonstrate that the proposed decentralized approach achieves comparable or higher localization accuracy than its centralized counterpart while reducing localization time by up to 34%. These findings highlight that decentralized graph-based RCL can provide a practical and efficient solution for anomaly diagnosis in resource-constrained edge environments.

cs.DC↗

arXiv2024

iAnomaly: A Toolkit for Generating Performance Anomaly Datasets in Edge-Cloud Integrated Computing Environments

Microservice architectures are increasingly used to modularize IoT applications and deploy them in distributed and heterogeneous edge computing environments. Over time, these microservice-based IoT applications are susceptible to performance anomalies caused by resource hogging (e.g., CPU or memory), resource contention, etc., which can negatively impact their Quality of Service and violate their Service Level Agreements. Existing research on performance anomaly detection in edge computing environments is limited primarily due to the absence of publicly available edge performance anomaly datasets or due to the lack of accessibility of real edge setups to generate necessary data. To address this gap, we propose iAnomaly: a full-system emulator equipped with open-source tools and fully automated dataset generation capabilities to generate labeled normal and anomaly data based on user-defined configurations. We also release a performance anomaly dataset generated using iAnomaly, which captures performance data for several microservice-based IoT applications with heterogeneous QoS and resource requirements while introducing a variety of anomalies. This dataset effectively represents the characteristics found in real edge environments, and the anomalous data in the dataset adheres to the required standards of a high-quality performance anomaly dataset.

cs.DC↗

arXiv2024

Efficient Training Approaches for Performance Anomaly Detection Models in Edge Computing Environments

Microservice architectures are increasingly used to modularize IoT applications and deploy them in distributed and heterogeneous edge computing environments. Over time, these microservice-based IoT applications are susceptible to performance anomalies caused by resource hogging (e.g., CPU or memory), resource contention, etc., which can negatively impact their Quality of Service and violate their Service Level Agreements. Existing research on performance anomaly detection for edge computing environments focuses on model training approaches that either achieve high accuracy at the expense of a time-consuming and resource-intensive training process or prioritize training efficiency at the cost of lower accuracy. To address this gap, while considering the resource constraints and the large number of devices in modern edge platforms, we propose two clustering-based model training approaches : (1) intra-cluster parameter transfer learning-based model training (ICPTL) and (2) cluster-level model training (CM). These approaches aim to find a trade-off between the training efficiency of anomaly detection models and their accuracy. We compared the models trained under ICPTL and CM to models trained for specific devices (most accurate, least efficient) and a single general model trained for all devices (least accurate, most efficient). Our findings show that the model accuracy of ICPTL is comparable to that of the model per device approach while requiring only 40% of the training time. In addition, CM further improves training efficiency by requiring 23% less training time and reducing the number of trained models by approximately 66% compared to ICPTL, yet achieving a higher accuracy than a single general model.

cs.DC↗