Sarcouncil Journal of Engineering and Computer Sciences

Abstract: Artificial Intelligence (AI) in combination with cloud computing has led to new performance, scalability and security challenges on the cloud network architecture that have never been seen before. A classically designed cloud-native deployment oftentimes misunderstands the high throughput, low latency, and elastic needs of AI workloads, especially when training an AI model at scale, and running real-time inference. In this paper we propose a set of ideas, framework to be able to optimize cloud networks with specific focus on optimizing cloud networks to run AI/ML workloads, and there are four broad challenges as we see it related to network design, and security, observability and operational resilience. By means of empirical experimentation and architectural analyses, the paper will show the major benefits of AI-aware interconnects, elastic network scaling, zero trust segmentation and automated observability at greatly improving cloud-native AI systems performance and security posture. Quantitative findings show total latency reduction of up to 83 percent, 200 percent increase in throughput, and a 70 percent reduction in time of operational downtime in the operating systems by adopting the use of AI-augmented telemetry and implemented auto-remediation models. Besides that, the use of advanced access control and encrypted model pipelines reduces the impact of the emerging threats, such as the theft of models and data exfiltration. These results are buttressed by practical implementations in the AWS, Azure, and GCP. Embracing the principles of AI in the networking layer and thinking of architecture-as-code philosophy, the proposed framework can be used to deploy AI applications in a new cloud ecosystem securely, scale, and highly resilient. The strategic recommendations are given at the end of the study and perspectives of future research in the field of quantum-safe AI networking and edge intelligence are defined.