CoNA : Dynamic Application Mapping for Congestion Reduction in Many-Core Systems 2012 IEEE 30th International Conference on Computer Design (ICCD) M. Fattah, M. Ramirez, M. Daneshtalab, P. Liljeberg, J. Plosila 1

Outline  Introduction  Mapping Problem and Evaluation Metrics  Contiguous Neighborhood Allocation Mapping  Experimental Setup  Results and Analysis  Conclusion 2

Outline  Introduction  Mapping Problem and Evaluation Metrics  Contiguous Neighborhood Allocation Mapping  Experimental Setup  Results and Analysis  Conclusion 3

Introduction  An efficient algorithm for run-time application mapping problem  Three novel contributions  First node selection  First task selection  Map the rest of tasks onto nearest neighborhood 4

Outline  Introduction  Mapping Problem and Evaluation Metrics  Contiguous Neighborhood Allocation Mapping  Experimental Setup  Results and Analysis  Conclusion 5

Mapping Problem and Evaluation Metrics  Applications  A p =TG(T, E) t i T e i,j E  Communication platform  AG(Ñ, L)  ñ i,j ={(r i,j, pe i,j )| ñ i,j Ñ, 0≤ i<M, 0≤ j<N}  Manhattan Distance : MD(ñ i,j, ñ m,n ) = (|i - m| + |j - n|)  Mapping function  map: T → Ñ, s.t. map(t i ) = ñ m,n ; ∀ t i ∈ T, ∃ n m,n ∈ Ñ 6

Evaluation Metrics  Packet latency  Average Manhattan Distance  Average Weighted Manhattan Distance 7

Evaluation Metrics (cont.)  Mapped Region Dispersion  Internal Congestion Ratio (ICR)  The number of edges using the same channel with respect to its total number of edges 8

Outline  Introduction  Mapping Problem and Evaluation Metrics  Contiguous Neighborhood Allocation Mapping  Experimental Setup  Results and Analysis  Conclusion 9

Contiguous Neighborhood Allocation Mapping (CoNA)  Three steps  First node selection  Choosing the first task of the application  Contiguous neighborhood allocation 10

CoNA (cont.) 11

CoNA (cont.)  First node selection  The nearest node to the central manager among the nodes with the largest number of available neighbors 12

CoNA (cont.)  Choosing the first task of the application  Selects the task with the largest number of edges  The most intensive communication 13

CoNA (cont.)  Contiguous neighborhood allocation  Task graph is traversed in the breadth-first order, paired with their predecessors is: {(t 1, t 4 ), (t 2, t 4 ), (t 5, t 4 ), (t 0, t 1 ), (t 3, t 2 )}  Select the one which fits in the smallest square with the first node 14

CoNA (cont.)  Contiguous neighborhood allocation  Task graph is traversed in the breadth-first order, paired with their predecessors is: {(t 1, t 4 ), (t 2, t 4 ), (t 5, t 4 ), (t 0, t 1 ), (t 3, t 2 )}  Select the one which fits in the smallest square with the first node 15

CoNA (cont.)  Contiguous neighborhood allocation  Task graph is traversed in the breadth-first order, paired with their predecessors is: {(t 1, t 4 ), (t 2, t 4 ), (t 5, t 4 ), (t 0, t 1 ), (t 3, t 2 )}  Select the one which fits in the smallest square with the first node 16

CoNA (cont.)  Contiguous neighborhood allocation  Task graph is traversed in the breadth-first order, paired with their predecessors is: {(t 1, t 4 ), (t 2, t 4 ), (t 5, t 4 ), (t 0, t 1 ), (t 3, t 2 )}  Select the one which fits in the smallest square with the first node 17

CoNA (cont.)  Contiguous neighborhood allocation  Task graph is traversed in the breadth-first order, paired with their predecessors is: {(t 1, t 4 ), (t 2, t 4 ), (t 5, t 4 ), (t 0, t 1 ), (t 3, t 2 )}  Select the one which fits in the smallest square with the first node 18

CoNA (cont.) 19

Outline  Introduction  Mapping Problem and Evaluation Metrics  Contiguous Neighborhood Allocation Mapping  Experimental Setup  Results and Analysis  Conclusion 20

Experimental Setup  NoC platform  Plasma processor  Local memory  DMA controller  Tra-NI interface  Central manager (CM)  The maximum number of applications that could be injected per second into the system is denoted as λ full 21

Experimental Setup (cont.)  Simulation  To extract packet latency  FPGA  To investigate CoNA time complexity  Xilinx ML605 22

Experimental Setup (cont.)  Application set  Task graphs are randomly generated (set1) using the Task graph generator  Number of nodes : 4 – 11  Weight of edges : 4 – 16 flits  The weights of applications edges are equally multiplied by 16 (set16) 23

Outline  Introduction  Mapping Problem and Evaluation Metrics  Contiguous Neighborhood Allocation Mapping  Experimental Setup  Results and Analysis  Conclusion 24

Results and Analysis  Packet latency evaluation  Time complexity evaluation 25

Packet latency evaluation 26

Packet latency evaluation (cont.) 27

Packet latency evaluation (cont.) 28

Packet latency evaluation (cont.) 29

Time complexity evaluation 30

Time complexity evaluation (cont.) 31

Outline  Introduction  Mapping Problem and Evaluation Metrics  Contiguous Neighborhood Allocation Mapping  Experimental Setup  Results and Analysis  Conclusion 32

Conclusion  An efficient run-time task allocation is proposed  Reduce internal and external congestions  Three novel contributions 33

Thank you ! 34