synthesis of Interconnection Networks - College of Engineering ...

2 downloads 540 Views 283KB Size Report
The use of random regular networks was motivated by their surprisingly good ... performance and fault-tolerance measures of interest ..... A graphical tool to fa-.
                                              

                                                 Æ                                                                                                                          !                       "                             #$     "                               "                           "   "       

  

                        

                    

                    

                            

      

          !                               



          "          

  

    #               

             

        

$        %                & '              () *+ ,  

        "     !           !                     

   

                    !   "          #       "        "   $  Æ" "   % &   '   % " & #  #  !  (  $&        &   )*  !% ½

                    

          %           "     

    

 

- !  

   

    !     

2       

        

   

   !   

,        

    

                

       Æ                     

6                 

    

             
?+

    (>4+      3 (>+ 9   (>@+   

 92   3 (@+

  4 

            4

;      

(A+    1  

        

        " 

  

 

     B   

  

   F 4G?              AG)    

7           

    !

   

   

 

6 !         

   

 

         C 

          

     

 !

     1                C        

3.1. Generating Random Regular Networks

 0     !   

      

 6

        $ 

&                                  

          

 

   "    

 

=   

  

    0     

 

  -          

    

          6

'(               !    '(     

C           #         

 % >            



  A  >@                                    

    "   $    =    # >     5EE9," :  @5?9'  

 

      @E4)      

                      

 

         

 >@         !               H    ! 

       !    

    " 

        

+   >      +  + =           +   D         

          

,+ +

! 

        





       

  !

-+ .+ + + + + + + ,+ +

     C 

% 

     D >                 

     D E

  



  

    **I           %                     ./             "  

  (>5+

3.2. The Filtering Process      ! 

       #               



      

 

 C 

 

            

 !   "   >     

    0    !        !      

              % ! 

     6           

               

  

  !   

 #

         

0.040 degree 8

0.030

Filter 1

0.025

degree 6

0.020

Filter 2

0.010

E2 C2 U

C=C1 C2 C 3

degree 3

0.005 0

50

100 150 200 Number of nodes

250

300

Filter 3

E3 C3 E C

Figure 1. Average runtime of the generation algorithm



E3

E2

E1

degree 4

0.015

0

E1 C1

Evaluation part Checking part

Figure 2. Sequential and Parallel Filtering

       

                - !                 

        !          !  

                          ! 

            !

U

Average Runtime (sec)

0.035



 ! 

         !            !          !             

!          

           !            

  !          ! 2      7     C               

      

2   

            %      !                     5I                       

    

!        !         Æ  

                  

   

  ! 

    

!        

         

    !                  

4.1. The Diameter Filter

  !                   

  J      

!      !  

        

   # @      

        

      !     

        

     !   

         " 

        



  

     

   

!        

      (4 >3+ 0  

  

   

  

          

          

           

 "    



   

# # 4       

       "  

         

   

       

     

        



 "  )  @5?    

 "  ,    92   

   92   

  3  ?     

1  92   !   

#  "   >EEE   

"  

   >

½ 

  

         

½   %    4   

 # 3      

      

   "        

 

           

 

       

         ! #

 !    $   # 5  

 4         

      3  



   3      

 !         

        

     # 3

    !     

       !   

10

  

1.0

degree 3

0.8

8

Diameter ( ∆ )

8

Diameter ( ∆ )

9 degree 4

7

degree 5

6

6

0.6

Min. Diameter count

4

0.4

2

0.2

5 degree 6

4 3

0

2 0

50

100

150 200 Number of nodes

250

300

Figure 3. Diameter of random regular networks

0

20

40

60 80 Number of nodes

100

120

0 140

Figure 5. Frequency of minimum diameter random networks of degree 3

14 # 4         

Diameter ( ∆ )

12

    "    

10

         

8

       

         

6

      

Shuffle Exchange Network Cube Connected Cycles Chordal Ring of degree 3 Moebius Graphs Random Networks of degree 3 Multi Tree Structures

4 2

 

      "                       9 (3+            (5 ?+

0 0

20

40

60

80 100 120 Number of Nodes

140

160

Figure 4. Diameters of different networks of degree 3

180

#            

 J

¾

   >    

½ %  #  % "  / *   0 !    "1   ¾  2  34 " $!           "$ 0  $    $  %  3 56

Fraction of the generated graphs

Min. Diameter

10

  

       

          



  

    >       

    

 

         

        

     $                   

    

   

      



             >EE   

2"  

;

;

" ?4   4   5   5 

'  K

=

       " ?4

>E

3

@

@

  4    # 3 #   

>5

4

@

3

        

@E

3

3

4

        C 

AE

3

5

?

    

3?4

4

5

?

          

53@

5

5

?

         

A4E

4

?

A

        

Table 1. Comparison of diameter between best known networks and the best of the random networks generated in our experiments

      

    # ?

4.2. The Scalability Filter

Before Scaling

After Scaling

1 0 0 1 0 1 00 11 00 11 00 11 00 11 00 11 00 11 00 11 00 11

11 00 00 11 00 11 00 11 00 11 00 11 00 11 00 11 00 11 0 1 0 1

New Node

2      

Edge removed

1 0 0 1 0 Designated node 1

   %       

Edge added

   "                          #                  

Figure 6. Example of edges that can be removed to accomodate a new node

         "         

         

     "     

          

      

     # A   

  =       

          

         "

 

    

    !      >EI  

     # 3 

          

         

         

   " 

         E E5



                         

                 

@

  

          

  >

4.3. The Fault-Tolerance Filter

 

    



 1 

       

  

=             9         

   

    

    

          

       C    

 





           

  

   6   

     

             

  0 !    

 J                   

         

  "     





      

 

0.8 Diameter (∆)

Fraction of input graphs

1

0.6 0.4 0.2 0

0

0.1

0.2 0.3 Diameter increase

0.4

0.5

 

   

   (>>+          "    

       0          

          0      " ?4   3  4          "                   

Diameter (∆)

        

   

64-node shuffle 64-node CCC 64-node chordal ring of degree 3 Moebius graph of order 6 64-node random network of degree 3 64-node MTS graph of degree 3 0

0.05

16 15 14 13 12 11 10 9 8 7 6 5 4

0

0.05 0.10

# )       

0.15 0.20 0.25 0.30 Probability of link failure

0.35

0.40

(b) Comparison among degree 4 networks

  3   4             3 

0.25

16 x 4 mesh torus 8 x 8 mesh torus 64-node chordal ring of degree 4 4-D wrapped butterfly 64-node random network of degree 4

      !

    4    

0.10 0.15 0.20 Probability of link failure

(a) Comparison among degree 3 networks

Figure 7. Cumulative frequency of the increase in diameter for random networks of size 64 and degree 3

  

15 14 13 12 11 10 9 8 7 6 5

Figure 8. Diameter vs. probability of link failure

     # *  >>               3    

          

       4 

         $

             

       

 %        

             

        

 

        

         

          

        

         

       ,    

          & 

!           

    

  

  !           ! 6

 

        

7    

           

      

        

    !   

 

       

Max. connected component size (χmax)

Average Distance (D)

6.5 6.0 5.5 5.0 64-node shuffle 64-node CCC 64-node chordal ring of degree 3 Moebius graph of order 6 64-node random network of degree 3 64-node MTS graph of degree 3

4.5 4.0 3.5

0

0.05

0.10 0.15 0.20 Probability of link failure

0.25

Probability of disconnection ( π d)

0.5

63 62.5 62 64-node shuffle 64-node CCC 64-node chordal ring of degree 3 Moebius graph of order 6 64-node random network of degree 3 64-node MTS network of degree 3

61.5 61 60.5 60

0

0.05

0.1

0.15

0.2

0.25

Figure 11. Maximum component size vs. probability of link failure

64-node MTS graph of degree 3 64-node random network of degree 3 Moebius graph of order 6 64-node chordal ring of degree 3 64-node CCC 64-node shuffle

0.6

63.5

Probability of link failure

Figure 9. Average node-pair distance vs. probability of link failure

0.7

64

!     

  

    !  L  " ?4   4    !  

     !       !       !    A          !

0.4

         



0.3



    !   J   !    !        

0.2

  

        (E E E @+ 7       >EEE    %

0.1



    !    

0

0

0.05

0.10 0.15 Probability of link failure

0.20

0.25

33I D33E                            

  !         

Figure 10. Probability of disconnection vs. probability of link failure

       5I 

            !





    > 5I D>5

  

   .  /              2                      

                     $   

                     

          

         !  

2 "        

          !

      Æ

%     

       

         

  !   

    !      

      !    

         

        

         1

-      

         

   ! 

    ! 



3    9 ; &  7  1

   # 

   

          !

#    

- !       

1  ! 9   $

   

  

  



& 

        

!          

/""'!'/ 50  /

"

     !     7        



 %                   

 

      M K     

    

   

     



  

  

   

 ! " #

 $ 

      %    &% & #% &

   

'!' 

(  $ 

)    * + , -  &  #

    /

( &   

0 

*  &   #

        

 %1

 

#   !$  .

!.

 $ "

 

0 

2 &  #

    

           

#  $!

 $ '

 

0   3  4  *  &   & %   #

.

6 7

  

"!/ 50  $

8      #   

 

   

//.!

/" 9   $/ $

 !

* 5 

       !

  "

    #  $  !!!  %      +))) : :    8(  $/ 

2 5   9 ; &  :   

   

 

 '!. 9 

 $ 

#