Numerical Evaluation of Network Latency and Throughput in Distributed Systems

Authors

Saira Akram, Muhammad Asim Akram, Fattah UR Rehman Jiangsu university science and technology, Cholistan university of veterinary and animal science, Preston university, Islamabad, Pakistan Author

Distributed Systems, Load Balancing, Network Latency, First Passage Time, Queueing Theory, Load-Sharing Window.

This research investigates the critical impact of network latency and throughput on the efficiency of load-sharing

algorithms in distributed computing systems. While traditional load-balancing strategies often rely on instantaneous

state information, they frequently fail to account for the stochastic nature of the Load-Sharing Window (LSW) the

duration during which a node remains in a specific state. This paper develops a mathematical framework using

M/M/1 queueing models to numerically evaluate the probability of successful job transfers under varying traffic

intensities (ρ) and communication delays.

Our analysis identifies two primary failure modes: Transfer in Vain (TIV), where the receiver becomes busy before

the job arrives, and Transfer Unnecessary (TU), where the sender clears its own backlog during the transfer process.

To mitigate these inefficiencies, we propose and evaluate a β -quantile decision rule that utilizes the probability

distribution of the LSW to filter out high-risk migrations. Numerical results demonstrate that in high-load scenarios

(ρ= 0.9), the quantile-aware approach improves effective system throughput by up to 25% compared to latency-blind

algorithms. The study concludes that incorporating the statistical confidence of the LSW into scheduling decisions is

essential for maintaining system stability and maximizing resource utility in high-latency distributed environments.