Cloud computing offers cost efficiency, flexibility, and scalability, but efficient resource managemen, especially dynamic scaling, remains a key challenge. This paper presents a predictive autoscaling strategy using transformer-based workload prediction model integrated with Kubernetes-based Event Driven Autoscaler (KEDA) to dynamically scale an ingress controller based on anticipated workloads. Trained on real-time data from Alibaba Cloud, the transformer model was deployed in a cloud-native Kubernetes environment. Performance was evaluated through load testing against standard autoscaling methods, including Horizontal Pod Autoscaler (HPA). Results show that the proposed model maintains stable response times and improves resource utilization, with further gains possible through hyperparameter tuning. The study demonstrates that transformer-based predictive autoscaling can reduce resource wastage and mitigate performance degradation during traffic surges, offering a promising direction for future cloud resource management solutions.
Anomaly detection is essential for ensuring the safe and efficient operation of industrial systems like smart grids. Smart grid stations handle sensitive data and are often hesitant to share it with third-party servers for centralized anomaly detection. Federated Learning (FL) offers a viable solution to this issue by enhancing anomaly detection in smart grids without compromising data privacy. We present a method for developing an unsupervised anomaly detection system using FL applied to a synthetic dataset that mimics a real-world smart grid system's behavior. We focus on utilizing FL's long short-term memory autoencoder in short, LSTM-AE for anomaly detection. However, there are concerns about potential privacy breaches in the FL system. Hence, to address this issue, we propose to integrate differential privacy (DP) with FL for anomaly detection by adding artificial noise to parameters at the client side before aggregation. This method ensures data privacy while maintaining the convergence of federated learning algorithms. Moreover, this research determines the optimal privacy level to balance noise scale and model accuracy. Our findings suggest a criterion for selecting the right privacy budget of DP based on the requirement of the system to provide good level of privacy in the system while maintaining the f1-score of FL-based anomaly detection system greater than 90%.
Latency, jitter, and reliability are three KPIs in QoS of modern edge and cloud computing systems. Their analytical insights can be well gained through queuing theory. In this paper, we develop a queuing model with two parallel heterogeneous servers with a randomly-initial-dispatch discipline. This model generalizes several existing models, yet simple enough as it can be configured by a couple of parameters. With this simplicity, we analytically prove the optimality of several standard measures of effectiveness for the latency and jitter.
In the realm of Database-as-a-Service (DBaaS), multi-tenancy has emerged as a key concept, enabling a single database instance to be shared among multiple tenants. Within a multi-tenant parallel DBMS, at any point, many tenants may submit SQL queries concurrently, each associated with a performance Service-Level Objective (SLO). Any violation to these SLOs requires the provider to pay fines. Given the instability of query workload in these systems, along with the need to satisfy the strict performance objectives, innovative query optimization strategies are required. These strategies should be well adapted to the specific needs of multi-tenant environments while ensuring profitability for the service provider. In this paper, we design a workload prediction management system (or Prediction Manager in short) inside the query optimizer. The main objective of this manager is to continuously predict the upcoming workload and make this information available to the optimizer. This enables the query optimizer, along with the performance SLOs, to decide when and which execution plan to select. Experimental results showed high efficiency in terms of the provider's long-term profit relatively to those in existing literature.
The rapid increase in data volume over the past years has given rise to big data science. Distributed File Systems (DFS) have become widely employed to handle this vast amount of data, such as Google File System and Hadoop Distributed File System (HDFS). The primary objective of a DFS is to ensure data availability and system reliability in the event of failure. Data availability and system reliability are achieved by replicating files across multiple locations, which, however, results in the consumption of storage space and other resources. The significance of these files varies based on their frequency of use within the system. As a result, specific files are deemed less critical and do not warrant extensive replication, as they hold little importance in the overall system. This paper presents a novel approach called "Optimizing HDFS Replication Policies with Predictive Categorization" for storage efficiency. This approach aims to minimize storage consumption while ensuring data availability and system reliability. Experimental results using the Spotify Hit Predictor Dataset (1960-2019) showcase significant improvements in storage utilization and throughput. This approach not only enhances system performance but also positions organizations for substantial cost efficiencies in managing their data-intensive workloads.
