Fluid Limits for Time-Varying Many-Server Queues with Finite Capacity
MINGRUI WANG, PRAKASH CHAKRABORTY · 2025
Read Fluid Limits for Time-Varying Many-Server Queues with Finite Capacity by MINGRUI WANG, PRAKASH CHAKRABORTY (2025) in the Tower of Babel, a free online library of public-domain works preserved in clean, searchable text. No sign-up. The full text is served as a plain-text file at a predictable URL, where text is present: https://antarctic-labs.com/tower-of-babel/arxiv-2511-07650.txt.
Introduction
Mingrui Wang and Prakash Chakraborty derive fluid limits for call-center style systems whose demand changes through the day. Many servers face arriving jobs with general service times, and zero waiting room means late arrivals are lost. Blocking happens instantly, making the limiting equations discontinuous.
A functional strong law shows the busy-server fraction follows a nonlinear Volterra integral equation. A careful solution concept makes that equation well-posed and yields time-varying acceptance odds without hand-waving. Finite waiting room versions get parallel treatment for busy, waiting, and lost fractions. Service operators gain exact large-scale laws for staffing systems that never sit still.
Catalog record
| CREATOR | MINGRUI WANG, PRAKASH CHAKRABORTY |
|---|---|
| YEAR | 2025 |
| COLLECTION | PAPERS |
| FORMAT | TXT |
| FILE SIZE | 129 KB |
| VERSION | 1.0 |
| LICENSE | http://creativecommons.org/licenses/by/4.0/ |
| RIGHTS | CC BY |
| SOURCE | arXiv |
Source: https://arxiv.org/abs/2511-07650
Browse the full library · About the Tower of Babel
Full text
The opening of the work, from the library's hosted text:
Fluid Limits for Time-Varying Many-Server Queues with Finite Capacity Thanks: P. Chakraborty is partially supported by the National Science Foundation under grant DMS-2153915 MINGRUI WANG, PRAKASH CHAKRABORTY Abstract . This paper develops fluid limits for nonstationary many-server loss systems with general service-time distributions. For the zero-buffer M t / G / n / n M_{t}/G/n/n queuing model, we prove a functional strong law of large numbers for the fraction of busy servers and characterize the limit by a nonlinear Volterra integral equation with discontinuous coefficients induced by instantaneous blocking. Well-posedness is established through an appropriate solution concept, yielding the time-varying acceptance probability without heuristic approximations. We then treat the finite-buffer M t / G / n / ( n + b n ) M_{t}/G/n/(n+b_{n}) regime, proving a functional strong law of large numbers for the triplet of fractions of busy servers, occupied buffers, and cumulative departures, whose limit satisfies a coupled system of three discontinuous Volterra equations capturing the interaction of service completions, buffer occupancy, and admission control at the capacity boundary. We establish well-posedness and convergence of the time-varying acceptance probability. Our theoretical results are supported by numerical simulations for both zero and finite-buffer regimes, illustrating the convergence of transient acceptance probabilities guaranteed by our theory. Finally, we use the fluid limits to derive optimal staffing and buffer-capacity for both time-varying loss systems. 1. Introduction Many modern service systems operate with limited capacity, meaning customers are turned away or lost when the system is full. Classic examples include telephone networks with a fixed number of trunk lines [ 21 , 11 , 18 ] , hospital or emergency units with limited beds [ 9 , 2 , 1 ] , wireless and optical networks with bandwidth and channel constraints [ 25 , 36 , 37 , 41 ] , emergency services like ambulances and self-driving cars [ 17 ] . These loss models, sometimes called Erlang loss systems, have been studied extensively under steady-state conditions. In fact, the famous Erlang-B formula [ 12 ] developed over a century ago for telephone traffic gives the steady-state blocking probability for an M/M/n/n queue and remains a cornerstone result in stationary loss models. Yet real-world systems are rarely stationary: arrival rates and service demands fluctuate over time, and service durations are not necessarily memoryless. As a result, steady-state measures often fail to capture short-term dynamics, leading to inefficient or unstable operational decisions. Nonstationary, non-Markovian loss systems such as M_{t}/G/n/n queues are significantly more challenging to analyze, and closed-form transient performance formulas are virtually impossible to obtain. This difficulty motivates the use of stochastic-process approximations for performance analysis, especially in many-server regimes where the number of servers n is large. Fluid limits or functional strong laws of large numbers (FSLLN) provide deterministic approximations to many-server queuing systems by tracking the scaled system state as n\to\infty . These limits reveal the macroscopic law of motion of complex stochastic systems. Foundational work such as [ 16 , 28 ] introduced asymptotic techniques for many-server systems and Markovian service networks. Subsequent research established fluid and diffusion limits under increasingly general conditions, including time-varying arrivals and non-exponential service times [ 33 , 20 , 26 , 27 , 42 ] . In contrast to these limit theorems, an extensive applied literature has developed practical approximations and staffing heuristics for time-varying service systems. Related work, including [ 19 , 14 , 15 , 38 , 40 ] , proposes pointwise-stationary (POS), modified-offered-load (MOL), and other transient approximations aimed at dynamic staffing, capacity planning, and transient performance evaluation. These studies underscore the need for rigorous transient characterizations that connect operational heuristics with asymptotic theory. 1.1. Overview of Approach and Key Insights. This paper develops a rigorous fluid-limit framework for analyzing time-varying many-server loss systems . Specifically, we study a sequence of systems with nonhomogeneous Poisson process (NHPP) arrivals and general service-time distributions, where both the number of servers and the arrival rate scale linearly with system size. The resulting limit is characterized by a nonlinear Volterra integral equation (VIE) that captures the transient evolution of the system’s occupancy and, crucially, yields the time-dependent blocking and acceptance probabilities in the large-scale regime. Our work builds upon [ 8 ] , which established a fluid limit for the nonstationary many-server M_{t}/G/n/n loss system using a semimartingale representation of the instantaneous acceptance mechanism. We enhance that framework by introducing a refined convergence proof based on the discontinuous Volterra equation methodology of [ 24 ] , ensuring well-posedness and uniqueness of the limit even under nonsmooth boundary dynamics. A distinguishing feature of our analysis is the emergence of nonlinear Volterra equations with discontinuous coefficients , induced by the instantaneous blocking constraint at full capacity. This structure departs sharply from classical Markovian formulations and provides a new analytic mechanism to capture threshold-type, transient blocking phenomena in nonstationary systems. The discontinuity is not merely a technical complication. It serves as the deterministic counterpart of the system’s stochastic acceptance barrier and encodes the operational behavior of loss systems under time-varying load. From a methodological standpoint, our results bridge three traditions in the study of nonstationary queues: (i) steady-state or quasi-stationary approximations such as the Erlang-B, PSA, and MOL methods [ 14 , 29 , 15 , 39 ] ; (ii) computational and moment-based approximation methods, including cumulant and truncated-ODE approaches for time-varying loss and many-server systems [ 31 , 30 , 19 , 38 , 40 ] ; and (iii) rigorous asymptotic limit theorems [ 16 , 28 , 33 ] . The fluid model derived here serves simultaneously as a limit theorem and a computational engine. It is a deterministic equation directly solvable by numerical methods, providing transient blocking probabilities without Monte Carlo simulations. This connection between rigorous scaling limits and practical performance computation strengthens the link between applied probability and operational analysis, particularly in time-dependent service environments such as healthcare scheduling, cloud service provisioning, and mobility-on-demand platforms. Accurate transient blocking or acceptance probabilities support dynamic staffing and admission-control decisions under fluctuating demand. Whereas traditional time-varying approximations assume local equilibrium, our limit provides a theoretically consistent foundation for approximating time-varying acceptance probabilities under nonstationary demand, which is central to time-dependent operations management. Although our analysis focuses on NHPP arrivals, the fluid-limit structure depends only on the arrival-rate trajectory rather than Poisson-specific properties. The same analytical framework extends naturally to renewal or Cox processes with time-dependent intensities. This generality implies that the derived acceptance probabilities provide accurate first-order approximations for a broad class of time-varying queuing systems, highlighting the structural robustness of the fluid-limit formulation. 1.2. Contributions. We establish functional strong laws of large numbers for nonstationary many-server loss systems under general service-time distributions. Both the zero-buffer ( M_{t}/G/n/n ) and finite-buffer ( M_{t}/G/n/(n+b_{n}) ) systems are analyzed in a common framework that scales the number of servers and the arrival rate proportionally with system size 1 1 1 our analysis readily extends to time-varying piecewise constant service and buffer capacities. However, for simplicity, we consider the case where both are constant. . The resulting limits are deterministic trajectories described by nonlinear Volterra integral equations (VIEs) that capture the transient evolution of the system occupancy and yield the associated time-dependent blocking and acceptance probabilities. (i) Zero-buffer