Search

Cloud computing has become the cornerstone of modern technology, propelling industries to unprecedented heights with its remarkable and recent advances. However, the fundamental challenge for cloud service providers is real-time workload prediction and management for optimal resource allocation. Cloud workloads are characterized by their heterogeneous, unpredictable, and fluctuating nature, making this task even more challenging. As a result of the remarkable achievements of deep learning (DL) algorithms across diverse fields, scholars have begun to embrace this approach to addressing such challenges. It has become the defacto standard for cloud workload prediction. Unfortunately, DL algorithms have been widely recognized for their vulnerability to adversarial examples, which poses a significant challenge to DL-based forecasting models. In this study, we utilize established white-box adversarial attack generation methods from the field of computer vision to construct adversarial cloud workload examples for four cutting-edge deep learning regression models, including Recurrent Neural Network (RNN), Long Short-Term Memory (LSTM), Gated Recurrent Unit (GRU), 1D Convolutional Neural Network (1D-CNN) and attention-based models. We evaluate our study with three widely recognized cloud benchmark datasets: Google trace, Alibaba trace, and Bitbrain. The findings of our analysis unequivocally indicate that DL-based cloud workload forecasting models are highly vulnerable to adversarial attacks. To the best of our knowledge, we are the first to conduct systematic research exploring the vulnerability of DL-based models for workload forecasting in the cloud data center, highlighting the inherent hazards to both security and cost-effectiveness in cloud data centers. By raising awareness of these vulnerabilities, we advocate the urgent development of robust defensive mechanisms to enhance the security of cloud workload forecasting in a constantly evolving technical landscape.

Edge computing has emerged as a promising computing paradigm that enables real-time data processing and analysis closer to the data source and boosts decision-making applications in a safe manner. On the other hand, the microservice is a new type of architecture that can be dynamically deployed, migrating across edge clouds on demand. Therefore, the combination of these two technologies can provide numerous benefits, including improved performance, reduced latency, and better resource utilization. In this paper, we present a thorough analysis of state-of-the-art research on the use of microservices in edge computing environments. We take into consideration several distinct microservice research directions, including coordination, orchestration, repositories, scheduling, autoscaling, deployment, resource management, and different security issues. Furthermore, we explore the potential applications of microservices in edge computing across various domains. Finally, the unsolved research issues and future directions of emerging trends in this area are also discussed.

In recent years, Knowledge Distillation has obtained a significant interest in mobile, edge, and IoT devices due to its ability to transfer knowledge from the large and complex teacher to the lightweight student network. Intuitively, Knowledge Distillation refers to forcing the student to mimic the teacher’s neuron responses to improve the generalization of the student by deploying the distillation losses as the regularization terms. However, the non-linearity of the hidden layers and the high dimensionality of the feature maps make the knowledge transfer a rigorous task. Though numerous methods have been proposed to transfer the teacher’s neuron responses in the form of diverse feature characteristics such as attention, contrastive representation, and so on, to the best of our knowledge, no prior works considered feature-level non-linearity during distillation. In this work, we ask, does feature-level non-linearity-based approaches can improve student performance? For investigating those concerns, we propose a novel knowledge distillation technique called the NeuRes (Neuron’s Responses) via distilling the Sparse Activation Maps (SAMs) to transfer the highly activated Neurons Responses to the student to enhance the representation capability. Proposed NeuRes selects the highly activated neuron responses that produce Sparse Activation Maps (SAMs) while transferring the knowledge based on activation normalization. Our proposed NeuRes also transfers the translation invariant features using auxiliary classifiers and augmented data to improve students’ generalization. The detailed ablation studies and extensive experiments on model compression, transferability, adversarial robustness, and few-shot learning verify that NeuRes outperforms state-of-the-art distillation techniques on the standard benchmark datasets.