Cong Wang

Anomaly detection in computational workflows is critical for ensuring system reliability and security. However, traditional rule-based methods struggle to detect novel anomalies. This paper leverages large language models (LLMs) for workflow anomaly detection by exploiting their ability to learn complex data patterns. Two approaches are investigated: 1) supervised fine-tuning (SFT), where pre-trained LLMs are fine-tuned on labeled data for sentence classification to identify anomalies, and 2) in-context learning (ICL) where prompts containing task descriptions and examples guide LLMs in few-shot anomaly detection without fine-tuning. The paper evaluates the performance, efficiency, generalization of SFT models, and explores zero-shot and few-shot ICL prompts and interpretability enhancement via chain-of-thought prompting. Experiments across multiple workflow datasets demonstrate the promising potential of LLMs for effective anomaly detection in complex executions.

Anomaly detection is the task of identifying abnormal behavior of a system. Anomaly detection in computational workflows is of special interest because of its wide implications in various domains such as cybersecurity, finance, and social networks. However, anomaly detection in computational workflows~(often modeled as graphs) is a relatively unexplored problem and poses distinct challenges. For instance, when anomaly detection is performed on graph data, the complex interdependency of nodes and edges, the heterogeneity of node attributes, and edge types must be accounted for. Although the use of graph neural networks can help capture complex inter-dependencies, the scarcity of labeled anomalous examples from workflow executions is still a significant challenge. To address this problem, we introduce an autoencoder-driven self-supervised learning~(SSL) approach that learns a summary statistic from unlabeled workflow data and estimates the normal behavior of the computational workflow in the latent space. In this approach, we combine generative and contrastive learning objectives to detect outliers in the summary statistics. We demonstrate that by estimating the distribution of normal behavior in the latent space, we can outperform state-of-the-art anomaly detection methods on our benchmark datasets.

Over the past few years, due to the boom of advances in image processing, edge computing, and wireless networking, unpiloted aerial vehicles, often referred to as drones, have become an important enabler to support a wide variety of scientific applications, ranging from environmental monitoring, disaster response, and wildfire monitoring to the survey of archaeological sites. In this article, we present the FlyNet platform, which extends an existing workflow management system to support and manage scientific workflows. FlyNet enables automated resource allocation, workflow instrumentation, and network service support to support researchers in their goal to analyze data for new scientific discoveries. In addition, FlyNet provides network services management to support quality of service for efficient data transport between edge devices, edge servers, and the cloud.

The Chameleon project developed a unique experimental testbed by adapting a mainstream cloud implementation to the needs of systems research community and thereby demonstrated that clouds can be configured to serve as a platform for this type research. More recently, the CloudBank project embarked on a mission of providing a conduit to commercial clouds for the systems research community that eliminates much of the complexity and some of the cost of using them for research. This creates an opportunity to explore running systems experiments in a combined setting, spanning both research and commercial clouds. In this paper, we present an extension to Chameleon for constructing controlled experiments across its resources and commercial clouds accessible via CloudBank, present a case study of an experiment running across such combined resources, and discuss the impact of using a combined research platform.

Identifying network anomalies is an important mea- sure to ensure reliability and quality of data transfers among facilities. Scientific workflows in particular heavily rely on good network performance to ensure their smooth executions. In this paper, we present a lightweight classifier system that is able to recognize anomalous TCP transfers. Using random forest trees and labeled data sets, we evaluate the classifier with real workflow transfers for ground truth data. Our studies reveal that various TCP congestion algorithms behave differently in anomalous conditions. We show that training classifiers on these separately can aid detection in network performance deterioration. Results reveal that our classifiers are able to better predict anomalous flows for TCP Reno and Hamilton compared to Cubic and BBR, due to the manner how their congestion control algorithms handle the anomalies.