Performance Slowdown Analysis

This repository contains the code and analysis for Assignment 2 of the course Software Performance and Scalability [CM0481]. The assignment focuses on studying and comparing the mean slowdown of various queuing disciplines including FCFS (First-Come-First-Serve, PS (Processor Sharing), and SRPT (Shortest Remaining Processing Time).

Introduction

The goal of the project is to compare the mean slowdown of three queuing disciplines: FCFS, PS, and SRPT. Slowdown, defined as the delay normalized to the job size, is used for comparison. Default values for the Pareto distribution shaping the policies are provided.

Final Considerations

All-Can-Win Theory Definition

The All-Can-Win theory states that for an M/G/1 system, if $\rho<\frac{1}{2}$, then every job prefers SRPT to PS in expectation. This property holds until $\rho<0.96$.

Plots Comparison

A comparison plot illustrates that SRPT consistently outperforms PS, validating the All-Can-Win theory. SRPT remains fair for all job sizes up to 400, with its performance consistently below PS.

Conclusions

The analysis confirms SRPT's superiority over PS, aligning with theoretical predictions. The results uphold the All-Can-Win theory, even for $\rho=0.7$ and $\rho=0.5$, satisfying the specified requirements.

The conclusions drawn from the analysis are summarized. It is observed that SRPT remains a fair discipline for job sizes up to 400. The All-Can-Win theory holds true for the given conditions ($\rho=0.7$), and it's inferred that the property is valid for $\rho=0.5$ as well.

Usage

1. Clone the repository.
2. Run plots.py to generate the comparison plot.
3. Review the analysis and conclusions in the README.md file.

Contributors

• Francesco Forcellato (875290)
• Gabriele Pilotto (902388)
• Pietro Visconti (885448)

Academic Year

2023 / 2024

For more details, refer to the code and documentation provided in this repository.