TABLE OF CONTENTS CHAPTER NO. TITLE PAGE NO. LIST OF TABLES LIST OF FIGURES LIST OF SYMBOLS AND ABBREVIATIONS

Transcription

1 vi TABLE OF CONTENTS CHAPTER TITLE PAGE ABSTRACT LIST OF TABLES LIST OF FIGURES LIST OF SYMBOLS AND ABBREVIATIONS iii viii x xiv 1 INTRODUCTION DISK SCHEDULING WINDOW-CONSTRAINED SCHEDULING CONTRIBUTION OF THE THESIS THESIS ORGANIZATION 7 2 RELATED WORK 8 3 DISK SCHEDULING INTRODUCTION PROBLEM FORMULATION PROPOSED SOLUTION STRATEGIES MOGA HGA Micro-GA IMPLEMENTATION AND RESULTS SUMMARY 45

2 vii CHAPTER TITLE PAGE 4 WINDOW-CONSTRAINED SCHEDULING FOR CPU TASKS INTRODUCTION PROBLEM FORMULATION PROPOSED SOLUTION STRATEGIES MOGA HGA Micro-GA IMPLEMENTATION AND RESULTS SUMMARY 72 5 WINDOW-CONSTRAINED SCHEDULING FOR MULTIMEDIA PACKETS INTRODUCTION PROBLEM FORMULATION PROPOSED SOLUTION STRATEGIES MOGA HGA Micro-GA IMPLEMENTATION AND RESULTS SUMMARY 85 6 CONCLUSION AND FUTURE WORK 86 APPENDIX 1 88 REFERENCES 93 LIST OF PUBLICATIONS 101 CURRICULUM VITAE 102

3 viii LIST OF TABLES TABLE TITLE PAGE 3.1 Example of coding scheme in MOGA A sample disk scheduling problem Task parameters to calculate fitness function Parameters and values Example Set A randomly generated job set Comparison of micro population sizes GA Parameters Example set of 4 processes Comparison of GAs Example packet set Parameters and Values Example packet set for which U min < Comparative results for 3 packets Example packet set for which U min = Comparative results for 5 packets Example packet set for showing the missed deadlines 83 A1.1 Population size vs. Fitness 88 A1.2 Micro population size vs. Time taken for calculation 88 A1.3 Pareto front 88 A1.4 Traditional Algorithms vs. GA based Approaches 89 A1.5 esize vs. % of window-constraint violation 89 A1.6 Elitism rate vs. Fitness value reached 89 A1.7 Elitism rate vs. Converging generation 89

4 ix A1.8 Comparison of % of missed deadlines 89 A1.9 Comparison of % of average service delay 90 A1.10 Comparison of % average context switch 90 A1.11 % of feasible schedules produced 90 A1.12 Pareto front 91 A1.13 Converging generation 91 A1.14 Comparison of delay 91 A1.15 Utilization vs. Missed deadlines 92

5 x LIST OF FIGURES FIGURE TITLE PAGE 3.1 FCFS SSTF SCAN C-SCAN LOOK C-LOOK Steps for MOGA Algorithm for Population Initialization in MOGA Algorithm for Coding scheme in MOGA Algorithm for applying Elitism in MOGA Procedure for CX in MOGA Procedure for modified CX in MOGA Swap mutation The Proposed hybrid Algorithm The structure of micro-ga Micro-pop selection Population size vs. Fitness Micro population size vs. Time taken for calculation Comparison of Elitism rate in MOGA Effect of esize in micro-ga Pareto front produced Traditional Algorithms vs. GA based Approaches Algorithm for generating the job sets The gene representation for a job instance 50

6 FIGURE TITLE PAGE 4.3 Algorithm for generating the micro-population Algorithm for number of job instances serviced Algorithm for finding the number of context switches Algorithm for finding the maximum service delay Algorithm for first form of Elitism Algorithm for third form of elitism Algorithm for second form of elitism esize vs. % of window-constraint violation Elitism rate vs. fitness value reached Elitism rate vs. converging generation Comparison of % of missed deadlines Comparison of % of average service delay Comparison of % average context switch Example job set for which VDS violates the windowconstraint though U min < Schedule produced by VDS when U min < Schedule produced by MOGA when U min < Schedule produced by HGA when U min < Schedule produced by Micro-GA when U min < Example job set for which VDS violates the windowconstraint though U min = Schedule produced by VDS when U min = Schedule produced by MOGA when U min = Schedule produced by HGA when U min = Schedule produced by Micro-GA when U min = % of feasible schedules produced Pareto front produced Result of MOGA 70 xi

7 FIGURE TITLE PAGE 4.29 Result of HGA Result of micro-ga Representation of packets Packet schedule produced by VDS when U min < Packet schedule produced by MOGA when U min < Packet schedule produced by HGA when U min < Packet schedule produced by micro-ga when U min < Packet schedule produced by VDS when U min = Packet schedule produced by MOGA when U min = Packet schedule produced by HGA when U min = Packet schedule produced by micro-ga when U min = Comparison of GA based approaches for convergence Comparison of delay Schedule produced by VDS for 3 packets Schedule produced by micro-ga for 3 packets Utilization vs. Missed deadlines 85 xii

8 xiii LIST OF SYMBOLS AND ABBREVIATIONS Symbols P - Boltzmann probability Abbreviations ACO - Ant Colony Optimization AS - Anticipatory Scheduler BNDP - Bicriteria Network Design Problem C-LOOK - Circular LOOK C-SCAN - Circular SCAN CAS - Cooperative Anticipatory Scheduler DWCS - Dynamic Window-Constrained Scheduling EDF - Earliest Deadline First Scheduling D-SCAN - Earliest-Deadline-SCAN EMOEA - Exploratory Multi-Objective Evolutionary Algorithm FD-SCAN - Feasible-Deadline-SCAN FCFS - First Come First Serve GA - Genetic Algorithm GSR - Global Seek-optimizing Real-time disk scheduling algorithm GSS - Group-Sweeping Scheduling GDPA - Guaranteed Dynamic Priority Assignment HGA - Hybrid Genetic Algorithm Micro-GA - Micro- Genetic Algorithm MOEA - Multi-objective Evolutionary Algorithm

9 xiv MOGA - Multi-objective Genetic Algorithm MOMGA - Multi-objective messy GA MOP - Multiobjective optimization problem NPGA - Niched Pareto Genetic Algorithm NSGA - Non dominated Sorting Genetic Algorithm P-DBCS - Packetized Dynamic Batch CoScheduling PAES - Pareto Archived Evolution Strategy PGA - Partitioned Genetic Algorithm pd-ga - period-based Genetic Algorithm pp-ga - Proportion-based Genetic Algorithm QoS - Quality of Service RAID - Redundant Array of Independent Disk SSTF - Shortest Seek Time First SA - Simulated Annealing SPEA - Strength Pareto Evolutionary Algorithm VEGA - Vector Evaluated Genetic Algorithm VDS PMX CX Virtual Deadline Scheduling Partially matched crossover Cycle crossover