Evolvable Fuzzy Hardware for Real-time Embedded Control in

Evolvable Fuzzy Hardware for Real-time Embedded Control in Packet Switching Ju Hui Li Meng Hiot Lim Qi Cao

Nanyang Technological University, Singapore Outline w Introduction to EHW w Evolvable Fuzzy Hardware (EFH) w Hardware Implementation (RFIC) w ATM Cell-Scheduling w EFH application on Cell-Scheduling w Simulation Results w Conclusion and Future Work 3/19/2018 2

Nanyang Technological University, Singapore Introduction to EHW w Definition n n Modify Autonomously w Classification Methods of Evolution l n Adaptation Scheme l n On-line, Off-line Evolutionary Granularity l 3/19/2018 Extrinsic, Intrinsic Transistor, Gate, Function Units 3

Nanyang Technological University, Singapore Introduction to EHW w Open Issues n n n 3/19/2018 On-line Adaptation Scalability Termination of Evolution 4

Nanyang Technological University, Singapore Evolvable Fuzzy Hardware w Evolvable Fuzzy Hardware (EFH) n Traditional Fuzzy Hardware Fuzzy Rule Set Designed by Experts l Considering The Whole Scenario l Fuzzy Rule Set Fixed l n EFH Fuzzy Rule Set Searched by GA l The Small Period Scenario l Fuzzy Rule Set may change l 3/19/2018 5

Nanyang Technological University, Singapore Evolvable Fuzzy Hardware w Architecture Evaluation 3/19/2018 6

Nanyang Technological University, Singapore Evolvable Fuzzy Hardware w Evolution Scheme n Training Data l n Pattern Prediction Search for a good but not optimal rule set The baseline is the working rule set l If no better chromosome can be found within the fixed generations, the working fuzzy rule set is deemed to be good enough l n 3/19/2018 Core rule set 7

Nanyang Technological University, Singapore Evolvable Fuzzy Hardware w RFIC (Reconfigurable Fuzzy Inference Chip) 3/19/2018 8

Nanyang Technological University, Singapore ATM Cell-Scheduling Problem w Problem Description n 3/19/2018 Class 1 Class 2 The capacity of Channels BUF 1 MP BUF 2 OUTBR 9

Nanyang Technological University, Singapore ATM Cell-Scheduling Problem w Quality of Service n n n 3/19/2018 Class 1 Cell Delay Cell Loss (Class 1 and Class 2) Balance of Cell Losses (Fairness) 10

Nanyang Technological University, Singapore ATM Cell-Scheduling Problem ---Available Schemes w FIFO w DWPS w Other Methods n n 3/19/2018 Round Robin Scheduling Generalized Processor Sharing 11

Nanyang Technological University, Singapore EFH Application on Cell-Scheduling Class 1 BUF 1 MP Class 2 BUF 2 Training buffer 1 (TB 1) Training buffer 2 (TB 2) Scheduling Model 3/19/2018 RFIC Evolution Module 12

Nanyang Technological University, Singapore EFH application on Cell-Scheduling w Training Data n n n Principle of “Locality” Using Past Period Cell Flow to Train EFH Search for a good but not optimal rule set. The baseline is the working rule set l If no better chromosome can be found within the fixed generations, the working fuzzy rule set is deemed to be good enough. l 3/19/2018 13

Nanyang Technological University, Singapore EFH application on Cell-Scheduling w Fuzzy Variables n n C 1=V 1/Vmax C 2=L 2/Lmax w Membership Functions 3/19/2018 14

Nanyang Technological University, Singapore EFH application on Cell-Scheduling w Core Rule Set n To Prevent From Adopting Very Poor Rule Set C 1 VS S M L VL T T T F T T T T F F F T T F F T 15 C 2 VS S M L 3/19/2018 VL

Nanyang Technological University, Singapore EFH application on Cell-Scheduling w Coding Scheme n 12222, 11122, 11111 w Fitness Function 3/19/2018 16

Nanyang Technological University, Singapore EFH application on Cell-Scheduling w GA Parameters n n n 3/19/2018 Generation Number=9 Evolution Cycle=2 Population Size=10 Elite Pool Size=2 Crossover Probability=0. 6 Mutation Probability=0. 05 17

Nanyang Technological University, Singapore EFH application on Cell-Scheduling w Simulation n Scenario 1 CBR is 155. 52 MHz l VBR is 155. 52 MHz l 2 Seconds l n Scenario 2 CBR is 100 MHz l VBR is from 55. 52 MHz to 155. 52 MHz l 2 Seconds l 3/19/2018 18

Nanyang Technological University, Singapore Simulation Results w Scenario 1 n 3/19/2018 19

Nanyang Technological University, Singapore 3/19/2018 20

Nanyang Technological University, Singapore 3/19/2018 21

Nanyang Technological University, Singapore Simulation Results w Scenario 2 n 3/19/2018 22

Nanyang Technological University, Singapore 3/19/2018 23

Nanyang Technological University, Singapore 3/19/2018 24

Nanyang Technological University, Singapore Simulation Results w Qo. S Tunability of EFH n n n 3/19/2018 Adjusting Qo. S by adjusting the value of λ. The smaller the λ, the smaller the class 1 delay and vice visa. The value of λ can be decided if the desired class 1 delay is decided. 25

Nanyang Technological University, Singapore Simulation Results w Tunability Simulation n 3/19/2018 26

Nanyang Technological University, Singapore 3/19/2018 27

Nanyang Technological University, Singapore 3/19/2018 28

Nanyang Technological University, Singapore 3/19/2018 29

Nanyang Technological University, Singapore 3/19/2018 30

Nanyang Technological University, Singapore Conclusions w The proposed EFH can be successfully applied on ATM cell scheduling w EFH can realize Intrinsic Evolution and Online adaptation. w It can trace the flow pattern and evolve an appropriate rule set. w It can achieve good Qo. S balance. w The achieved Qo. S can be adjusted conveniently 3/19/2018 31

Nanyang Technological University, Singapore Conclusions w (E) The result is equal to or better than the most recent human-created solution to a long-standing problem for which there has been a succession of increasingly better human-created solutions. w (F) The result is equal to or better than a result that was considered an achievement in its field at the time it was first discovered. w (G) The result solves a problem of indisputable difficulty in its field. 3/19/2018 32