🧮 Lab 3 — Introduction to Queuing Theory & Java Modelling Tools (JMT)

Duration: ~1 hour
Environment: Windows, Linux, or macOS (JMT is Java-based)

🎯 Learning Objectives

• Understand the basic components of a queuing system (arrivals, service, queues, customers).
• Learn to use the Java Modelling Tools (JMT) suite.
• Build, simulate, and analyze simple queuing networks.
• Compare system metrics: utilization, throughput, average queue length, response time.

⚙️ Setup Instructions

• Download JMT: https://jmt.sourceforge.net/
• Requires Java (≥ JDK 8)
• Unzip and run JSIMgraph.jar (for simulation models)

Tools Overview

🧩 Part 0 — Quick Recap: Queuing Theory Basics

Key Symbols:

• λ (lambda) = arrival rate (customers per second)
• μ (mu) = service rate (customers served per second)
• ρ (rho) = utilization = λ / μ

For M/M/1 queue:

• Average number in system: L = ρ / (1 − ρ)
• Average waiting time: W = 1 / (μ − λ)

🧩 Example 1 — RAID-0 style Parallel I/O (Fork–Join)

A request is split across k disks and the reply is sent after all parts finish (synchronous join). RAID (Redundant Array of Independent Disks) is a data storage technology that combines multiple physical disks into a single logical unit to improve performance, fault tolerance, or both. Different RAID levels (e.g., RAID 0, 1, 5) determine how data is split, mirrored, or parity-protected across the disks.

🧠 Learning outcomes

By the end of this exercise, students should:

• Understand the concept of fork–join synchronization in queuing networks.
• Observe that variability (heavy tail) in response time increases with more parallel disks.

⚙️ Model overview

Step-By-Step Instructions

Step 1. Create a new model

• Open JSIMgraph → click File → New Model.
• From the toolbar, drag a Source, Fork, Join, k Queueing Stations, and a Sink node onto the canvas.
• Connect them in this order:
• Source → Fork → (Disk1, Disk2, Disk3, …) → Join → Sink

Step 2. Create a Class

• Define distribution (e.g., exponentital, hperexponential etc). Choose λ, mean.

Step 3. Define Performance Indeces (what you want to measure)

Step 4. Run the simulation

🧩 Example 2 — Load Balancer Policies per Class

A router (LB) dispatches to a pool of servers. Compare Random, Round-Robin, and Shortest-Queue (SQ). Use two customer classes with different SLAs.

⚙️ Model overview

Step-By-Step Instructions

Step 1. Create a new model

• Open JSIMgraph → click File → New Model.
• From the toolbar, drag a Source, Router, k Queueing Stations, and a Sink node onto the canvas.
• Connect them in this order:
• Source → Router → (Server1, Server1, Server1, …) → Sink

Step 2. Create Classes representing different type of requests

• Define distribution (e.g., exponentital). Choose λ, mean.

Step 3. Define Routing Strategies for each class

Step 3. Define Performance Indeces (what you want to measure)

Step 4. Run the simulation

🧩 Example - Middleware Node with Prioritization and Network Link

🎯 Goal

In this experiment, we model a middleware system that receives messages of two types—High-priority (alerts) and Low-priority (telemetry)—and forwards them across a shared network link.
We want to study how priority scheduling and message filtering affect throughput and response time for each class.

• Processing: models message parsing or validation
• ClassSwitch: filters 20% of Low messages
• Prioritization: optional lightweight classification stage
• NetworkLink: single-server queue representing the network; uses class-based priority scheduling
• Sink: collects transmitted messages
• DropSink: collects filtered-out Low messages

⚙️ Model Configuration

📊 Performance Indices to Collect

For both classes and the system:

• Throughput [msg/s]
• Response time [s]
• Utilization [%] (of Processing and NetworkLink)
• Number of customers [msg] in queue and in service
• Drop rate [%] for Low messages (from DropSink throughput)

🔬 Experiments to Perform

1. Baseline scenario

   • λ_H = 2/s, λ_L = 8/s
   • Measure throughput and response times.
   • Observe how High messages experience consistently low delay.

2. Increased load

   • Increase both arrival rates (e.g., λ_H = 5/s, λ_L = 20/s).
   • Observe queue buildup at NetworkLink.
   • High messages still get through quickly; Low messages queue longer or drop.

3. Remove filtering

   • Change Low’s filtering probability to 1.0 (no drops).
   • Compare new response times—expect larger delays due to congestion.

⚙️ Example 5 — Parallel Multicore and Multithreaded Server

🎯 Goal

In this advanced example, we model a parallel server that executes multithreaded tasks on multiple CPU cores, all sharing a common memory bus.
This setup represents a multicore processor where threads run concurrently but contend for shared memory bandwidth.
The goal is to evaluate the scalability of the system and observe how resource contention limits performance.

🧩 System Architecture

Tasks → Fork (Create threads) → ([Core 1] [Core 2] [Core 3] [Core 4]) → MemoryBus →Join

Each job (user request) is decomposed into multiple threads, which are processed on different cores.
All threads share a single MemoryBus before synchronizing at the Join node.

⚙️ Model Configuration

🧠 Workload

🧩 Routing Configuration