Queuing Theory and Performance Modeling: Erlang, Little's Law, and the Pepper Language

Overview

This talk introduces the core ideas of performance modeling by focusing on resources and queues. It explains how simple abstractions like arrival rate and service time feed into meaningful metrics such as throughput, response time, and utilization, using Erlang’s formula and Little’s Law as cornerstones. The discussion then extends to networks of queues and to more expressive modeling frameworks that handle concurrency.

Takeaways

Readers will gain intuition for when queues form, how to quantify system performance, and how these ideas apply to modern technologies including AI systems and bio-inspired applications. The talk also highlights the tension between user needs and resource constraints faced by system operators.

Introduction to Queues and Resources

The presentation frames everyday experiences with queues as a lens on resource constraints in systems. It emphasizes that queues arise when demand for a resource exceeds supply and that performance modeling helps design better systems by analyzing this dynamic.

Erlang and Little’s Law

The speaker walks through Erlang's loss formula, showing how the workload E equals arrival rate multiplied by average duration, and how this drives the probability of blocking in a finite resource pool. The narrative then introduces Little's Law, which relates the average number in the system, the arrival rate, and the average time a customer spends in the system. The talk explains how these relationships hold under broad conditions and why they are foundational for capacity planning.

Observing and Measuring Queues

Key performance measures are defined: arrival rate, throughput, utilization, mean service time, and blocking probability. The talk discusses how to observe a system over a time horizon T to estimate these quantities and apply them to real world scenarios such as post offices, airports, and transportation networks.

From Queues to Networks

Moving beyond single queues, the talk describes queuing networks where multiple resources interact. Routing probabilities determine how customers move between queues, enabling analysis of more complex systems while preserving the queuing framework.

Two Modeling Paradigms

Historically, queuing networks provided elegant, compositional models. With higher interconnectivity and concurrency, the talk introduces process algebras as a modern approach. These algebras describe components such as customers and resources and can be compiled into Markov chains, yielding solvable mathematical models for performance metrics.

Applications and Evolution

The speaker shares diverse applications from freight canal optimization and cellular biology to cellular networks and 3G/4G/5G telecommunications. The narrative emphasizes that performance is about speed and responsiveness, not only correctness, and it shows how resource-aware modeling informs design decisions across domains.

AI and Resource Concerns

A forward looking section discusses the resource intensity of large language models and AI systems, arguing for performance modeling to guide efficient design and deployment. The talk closes by underscoring the value of robust mathematical models in understanding and managing resource use in modern technologies.