Enhancing queue operation efficiency under peak server load in parallel architectures with chunking authentication

Abstract

Relevance. In modern computing systems, operation queues and multi-server architectures play a central role in managing resource-intensive, heterogeneous workloads, such as artificial intelligence training, audio/video conversion, and industrial equipment optimization via computational modeling. Traditional single-processor systems exhibit inherent bottlenecks, including sequential execution delays and inefficient resource utilization, whereas multi-server distributions enable parallel processing, albeit at elevated energy costs. Prior work introduced a parallelization method for dust particle flow modeling, yielding computational time reductions. however, escalating server rental expenses and environmental imperatives require energy-focused improvements without losing speed. Aim and objectives. This study modernizes the Dispatcher-Deployer-Executor architecture, emphasizing real-time energy optimization in multi-processor environments. A review of existing studies highlights problems with predictive methods, such as using machine learning to predict needs, and reactive methods, such as scaling based on processor use, especially for unpredictable and varied tasks. Energy conservation via idle server sleep modes remains underexplored in such contexts. The primary objective is to minimize average energy consumption over operational periods while preserving task throughput and latency constraints. Methods used. The study employs a combined theoretical and empirical approach. In the improved model and method, support for power-saving modes was added with the following states: Active with full power, Waiting with reduced power, Hibernation with minimal power and memory disk-buffering, and Maintenance with zero power for multiple servers. Total energy consumption is modeled to be minimized subject to execution time bounds. State transitions are governed by idle thresholds, with activation criteria incorporating task execution time, queue wait, and transition latency. The improvements include “wait-for-executor-release” logic, where queued tasks defer activation, and dual modes: disk-buffering for fault-tolerant, asynchronous execution versus direct submission to available executors. Security enhancements encompass chunked cryptographic message authentication, protected transport-layer communication, and dynamic mutual authentication with certificate pinning and ephemeral certificates. Implementation leverages Rust’s asynchronous framework, with energy metrics via hardware interfaces. Results. To exemplify queued operations, numerical simulations computed trajectories of dust particles in gas flows. Differential equations of particle motion that account for viscous friction, inertia, and gravity forces were integrated using the Runge-Kutta-Fehlberg method. The calculation results are used for optimization of pipes and channels of energy equipment. Experiments on a multi-server cluster simulated heterogeneous loads. Baseline full-active mode was compared to optimized states, demonstrating significant energy reductions while maintaining performance. Buffering mode enhanced fault tolerance; submission mode improved latency. Conclusions. In conclusion, the enhanced Dispatcher-Deployer-Executor framework advances sustainable distributed computing, balancing reliability, speed, and energy efficiency for heterogeneous tasks. Future extensions may incorporate predictive analytics for proactive scaling and edge-cloud hybridization.