Optimization of Dispatcher-Deployer-Executor for short-term resource leasing in multiserver systems

Abstract

In modern distributed computing systems, architectures incorporating multiple servers and operation queues facilitate the management of heterogeneous resource-intensive workloads, such as artificial intelligence training, media file transcoding, and computational modeling for industrial optimization. Traditional single-processor systems are characterized by sequential constraints and inefficient resource allocation, whereas parallelism in multiserver configurations results in elevated energy consumption. The present study enhances the Dispatcher-Deployer-Executor (DDE) architecture for real-time resource consumption optimization in multiprocessor environments. The objective is to improve operation execution speed and minimize execution costs in scenarios requiring limited computational volumes and maximum parallelism. Through theoretical modeling and empirical verification, the extended DDE introduces a container mode for Deployer/Executor components, which operate within containers, enabling orchestration for short-term leasing via standard tools (e.g., Docker and Kubernetes). This implements dynamic resource scaling and efficient application of the model and method in cloud environments, with the primary advantage of rapid access to highperformance computations accompanied by substantial savings, as it eliminates the need for server acquisition or long-term leasing for restricted calculation volumes. The main application scenario is tasks oriented towards the central processing unit (CPU), as containers are optimized for CPU-bound computations, such as calculating the trajectories of dust particles through numerical solution of differential equations of motion. The results indicate a significant reduction in resource consumption and financial costs compared to the baseline mode while maintaining performance; quick rental of containers provides an advantage in cost savings and flexibility. The extended DDE contributes to sustainable computations, balancing efficiency, reliability, and speed for various tasks. Future extensions may integrate the use of standby mode and hibernation to reduce energy consumption.