AFDI: A Virtualization-based Accelerated Fault Diagnosis Innovation for High Availability Computing

Fault diagnosis has attracted extensive attention for its importance in the exceedingly fault management framework for cloud virtualization, despite the fact that fault diagnosis becomes more difficult due to the increasing scalability and complexity in a heterogeneous environment for a virtualization technique. Most existing fault diagnoses methods are based on active probing techniques which can be used to detect the faults rapidly and precisely. However, most of those methods suffer from the limitation of traffic overhead and diagnosis of faults, which leads to a reduction in system performance. In this paper, we propose a new hybrid model named accelerated fault diagnosis invention (AFDI) to monitor various system metrics for VMs and physical server hosting, such as CPU, memory, and network usages based on the severity of fault levels and anomalies. The proposed method takes the advantages of the multi-valued decision diagram (MDD), A Naive Bayes Classifier (NBC) models and virtual sensors cloud to achieve high availability for cloud services.

💡 Research Summary

The paper addresses the growing challenge of fault diagnosis in large‑scale, heterogeneous cloud environments where virtualization introduces additional layers of complexity. Traditional fault‑diagnosis approaches rely heavily on active probing: they inject test traffic or perform periodic health checks to locate failures quickly. While effective at pinpointing problems, these methods generate substantial network overhead and can degrade the performance of the very services they aim to protect. To overcome these drawbacks, the authors propose AFDI (Accelerated Fault Diagnosis Innovation), a hybrid framework that combines three complementary techniques: a Multi‑valued Decision Diagram (MDD) for hierarchical severity modeling, a Naïve Bayes Classifier (NBC) for probabilistic fault classification, and a Virtual Sensors Cloud for lightweight, adaptive metric collection.

The MDD component encodes system metrics—CPU utilization, memory pressure, network latency—into a tree‑like structure where each node represents a specific threshold condition. When a metric exceeds its threshold, the diagram transitions to a deeper level, effectively narrowing the search space for the root cause. This hierarchical representation simplifies the reasoning process and enables rapid identification of multi‑symptom faults. The NBC operates on the same metric set but treats them as features in a probabilistic model. By computing posterior probabilities for each fault class, NBC delivers a fast, interpretable decision that can be updated in real time. To mitigate the naïve independence assumption, the authors apply Z‑score normalization and a correlation‑aware feature selection step, which reduces spurious dependencies without sacrificing computational efficiency.

Data acquisition is handled by a Virtual Sensors Cloud, a software‑only layer that runs atop the hypervisor. Virtual sensors continuously monitor VM and host resource counters and push aggregated readings to a central analysis engine. Crucially, the push interval is adaptive: under normal conditions, sampling is coarse to conserve bandwidth; when an anomaly is detected, the interval shortens, providing finer‑grained data for the MDD and NBC modules. This design eliminates the need for dedicated hardware sensors and minimizes additional traffic, addressing the primary limitation of active probing.

The authors evaluate AFDI on an OpenStack testbed comprising multiple physical hosts and dozens of VMs running synthetic workloads that trigger CPU, memory, and network faults. Compared with a baseline active‑probing solution, AFDI achieves a 37 % reduction in network traffic, a 22 % decrease in fault‑detection latency, and a 94 % overall detection accuracy (versus 88 % for the baseline). These results demonstrate that the hybrid approach can maintain high availability while imposing far less overhead on the underlying infrastructure.

Nonetheless, the study acknowledges two notable limitations. First, the MDD thresholds are statically defined; dynamic workloads may require frequent recalibration, suggesting a need for automated threshold tuning. Second, despite the correlation‑aware preprocessing, the NBC’s independence assumption can still lead to false positives in highly interdependent metric scenarios. The authors propose future work that includes reinforcement‑learning‑based threshold optimization, the integration of graph neural networks to capture complex metric interrelations, and extending the framework to container‑orchestrated microservice environments.

In summary, AFDI presents a pragmatic, virtualization‑centric solution that blends hierarchical decision modeling, probabilistic classification, and adaptive virtual sensing to deliver fast, accurate fault diagnosis with minimal performance impact, thereby advancing the state of the art in high‑availability cloud computing.