Adaptive NLP-Driven Cloud–Edge Framework for Latency-Critical and Secure Applications

Abstract

A typical cloud resource management strategy is based on established policies and infrastructure-level metrics, but it may be inadequate to deal with latency sensitivity and privacy concerns in distributed application environments. The paper presents here an Adaptive NLP-Driven Cloud - Edge Framework combining transformer - based semantic workload analysis and intelligent resource orchestration. The framework workflow harvests work load properties from textual input, like logs, service requests and system alerts, such as latency sensitivity, computational demand, task priority and privacy requirement. A new algorithm, called Behavior-Aware Adaptive Resource Allocation (BAARA), is proposed to dynamically manage workloads based on workload requirements and runtime conditions between edge nodes, private cloud infrastructures and public cloud platforms. The applications on semantic workload interpretation and classification are using transformer-based models such as RoBERTa and DistilBERT. The hybrid cloud-edge simulation environment was used, consisting of 50,000 synthetic workload samples, for experimental evaluation. Experimental results shows that the proposed framework has reduced average response latency from 415 to 70 ms, and significantly reduced resource utilization until around 89% compared with the traditional scheduling, and also reduces the risk to privacy attack. A semantic workload analysis is coupled with an adaptive cloud-edge orchestration in order to achieve intelligent and context-aware resource allocation for customer latency-critical and secure distributed applications.