'interconnection network'에 해당되는 글 3건

Progress Report (9/15)

Lab.work 2009. 9. 15. 15:24
■ THIS WEEK  (9/8~9/14)

□ CS 710 project

I compared the cpi of IDEAL, MESH, PT_TO_PT.
Where PT_TO_TO is the topology that garnet offers defaultly (it is almost ideal).
MESH is the 4x4 mesh topology, 
IDEAL uses the same link latency and config with MESH but connected in a p2p manner.

For 3 application: barnes, cont-ocean, noncont-ocean,
The cpi result is PT_TO_PT < IDEAL < MESH
But the gap between them is not so big.
[mesh_vs_ideal.xlsx]

TODO: i think that the network topology does not seriously affect to the performance in splash2 applications
We might need to test specjbb or other and compare the results.

□ Booksim

I plotted 3-D graph of accepted packets for each nodes, but couldn’t find the specific pattern.
[booksim_hotspot.xlsx]

TODO: I want to fix the trafficmanager code to inject the same amount of packets for voq and no-voq.
and then, compare the results.

□ Phoenix

I changed the wordcount_map function to count all words.
(before, it only counted the words which consist of a-z)

The speedup pattern of map phase is not changed,
But the reduce time hugely increases.
From the result, the reduce processing time depends on the amount of words.

Actually, 
For the mapreduce version of wordcount, map threads store all words to the intermediate queue,
after that, reduce threads read words from the intermediate queue.
For the pthreads version, there is no intermediate queue between map and reduce. 
So, every word passes map and reduce at a time.
[test_on_gems.xlsx]

TODO: I think I need to check the size of intermediate queue, 
and check the latency of store, read words.

'Lab.work' 카테고리의 다른 글

Weekly report (10/5)  (0) 2009.10.12
Weekly report (9/28)  (0) 2009.09.29
Progress Report (9/8)  (0) 2009.09.08
PROGRESS REPORT (9/1)  (0) 2009.09.01
PROGRESS REPORT (8/26)  (0) 2009.08.26
블로그 이미지

민둥

,
■ Random traffic
모든 source는 모든 destination으로 1/N traffic
load balance가 매우 좋음 (매우 안좋은 topology나 routing의 경우에도 random traffic에서는 좋게 보일 수 있음)

■ Permutation
각각의 source는 모든 traffic을 하나의 destination으로 보냄

   □ Bit permutation 

Bitcomp

Bitrev

Bitrotation

Shuffle

Transpose

0000

1111(15)

0000(0)

0000(0)

0000(0)

0000(0)

0001

1110(14)

1000(8)

1000(8)

0010(2)

0100(4)

0010

1101(13)

0100(4)

0001(1)

0100(4)

1000(8)

0011

1100(12)

1100(12)

1001(9)

0110(6)

1100(12)

0100

1011(11)

0010(2)

0010(2)

1000(8)

0001(1)

0101

1010(10)

1010(10)

1010(10)

1010(10)

0101(5)

0110

1001(9)

0110(6)

0011(3)

1100(12)

1001(9)

0111

1000(8)

1110(14)

1011(11)

1110(14)

1101(13)

1000

0111(7)

0001(1)

0100(4)

0001(1)

0010(2)

1001

0110(6)

1001(9)

1100(12)

0011(3)

0110(6)

1010

0101(5)

0101(5)

0101(5)

0101(5)

1010(10)

1011

0100(4)

1101(13)

1101(13)

0111(7)

1110(14)

1100

0011(3)

0011(3)

0110(6)

1001(9)

0011(3)

1101

0010(2)

1011(11)

1110(14)

1011(11)

0111(7)

1110

0001(1)

0111(7)

0111(7)

1101(13)

1011(11)

1111

0000(0)

1111(15)

1111(15)

1111(15)

1111(15)


        ㅁ Bit complement : D_i = ~S_i
        ㅁ Bit reverse : D_i = S_b-i-1 (4x4 mesh에서 b=16)
           
[bitcomp]

[bitcomp]

[bitrev]

[bitrev]


        ㅁ Bit rotation : D_i = S_i+1 mod b
        ㅁ Shuffle : D_i = S_i-1 mod b
        ㅁ Transpose : D_i = S_i+b/2 mod b
          
[bitrot]

[bitrot]

[shuffle]

[shuffle]

[shuffle]

[transpose]



   □ Digit permutations
        ㅁ Tornado : D_x = S_x+(k/2-1)mod k (4x4 mesh에서 k=4)
             
[tornado]

[tornado]

        ㅁ Neighbor : D_x = S_x + 1 mod k
             
[neighbor]

[neighbor]



'Architecture' 카테고리의 다른 글

Seminar - Dec 28 (Mon)  (0) 2009.12.29
GARNET  (0) 2009.09.15
Virtual-Channel Flow Control  (0) 2009.07.08
Interconnection Network Topologies  (0) 2009.07.01
PARSEC vs. SPLASH-2  (0) 2009.06.16
블로그 이미지

민둥

,
□ Butterfly

(2-ary 4-fly)

k-ary n-fly Butterfly Network:
An n stage of radix k switches butterfly network. 
Such a network is composed of N=kn source/destination terminal nodes.
n stages of kn-1 switching nodes each with k inputs and k outputs.
All (n+1)N channels are unidirectional. 

□ Mesh
A k-ary n-mesh (mesh) topology


□ Cmesh (Concentrated Mesh)
Concentrated mesh topology. the 'c' determines the concentration



□ Fat tree
The links in a fat-tree become "fatter" as one moves up the tree towards the root.

□ Torus
A k-ary n-cube (torus) topology
A closed surface defined as the product of two circles


□ Quad Tree
A quadtree is a tree data structure in which each internal node has up to four children. 
Quadtrees are most often used to partition a two dimensional space by recursively subdividing it into four quadrants or regions.


□ Flattened Butterfly
"Flattened Butterfly : A Cost-Efficient Topology for High-Radix Networks" ISCA 2007

□ CMO (Concentrated Multi-dimensional Octagon)
Created by Sang Kyun Kim and Wook-Jin Chung from EE382C

□ MECS (Multidrop Express Channels)
"Scalable On-Chip Interconnect Topologies"


'Architecture' 카테고리의 다른 글

Network traffic patterns  (0) 2009.08.02
Virtual-Channel Flow Control  (0) 2009.07.08
PARSEC vs. SPLASH-2  (0) 2009.06.16
The PARSEC Benchmark Suite  (0) 2009.06.05
The SPLASH-2 Programs  (0) 2009.06.05
블로그 이미지

민둥

,