Skip to content

Latest commit

 

History

History
293 lines (265 loc) · 4.22 KB

performance-study.md

File metadata and controls

293 lines (265 loc) · 4.22 KB

Performance Testing

The following table shows the performance of the program across different versions and configurations. There are multiple tables, one for each machine used for testing.

NYCK-DESKTOP

Version 1000 64 0.4 0 200 128 .5 1000 10 512 .4 0 3 1024 .4 100
Serial 12.4s 22.5s
OMP 3.0s 12.0s 15.5s
MPI

NYCK-LAPTOP

Version 1000 64 0.4 0 200 128 .5 1000 10 512 .4 0 3 1024 .4 100
Serial 31.5s
OMP 22.5s 38.3s 132.6s 337.4s
MPI

Optimization Study

We will be studying how the implementations can be optimized, looking into Big O complexity, number of iterations and actual assembly instructions; and then comparing the performance of the different implementations to prove the hypotheses.

Serial

This is the basis of OMP implementation too, so figuring out the optimizations for this will help in the OMP implementation too.

Swap buffer

In serial, swap buffer is consistently around 2% faster than without it.

Buffer Swap Strategy

Serial

  • Without Buffer Swap: 356.7s
  • With Buffer Swap: 358.2s

Sample Instances

life3d 1000 64 0.4 0
life3d 200 128 .5 1000
life3d 10 512 .4 0
life3d 3 1024 .4 100

$ life3d 1000 64 0.4 0
1 81443 62
2 24563 24
3 20080 1
4 19016 1
5 17576 1
6 16905 1
7 15793 1
8 15174 1
9 14807 1

$ life3d 200 128 .5 1000
1 585667 87
2 198117 20
3 123360 6
4 117152 0
5 116181 0
6 116832 0
7 116421 0
8 116559 0
9 116344 0

$ life3d 10 512 .4 0
1 19157196 9
2 11835190 9
3 10389985 1
4 9659849 1
5 9049681 1
6 8563492 1
7 8146694 1
8 7824521 1
9 7580102 1

$ life3d 3 1024 .4 100
1 99923786 1
2 90413714 1
3 83137654 1
4 77287897 1
5 72448825 1
6 68444736 1
7 65198270 1
8 62633412 1
9 60611199 1

Species count increment inside loop with reduction:

14.7s
1 585667 87
2 198117 20
3 123360 6
4 117152 0
5 116181 0
6 116832 0
7 116421 0
8 116559 0
9 116344 0
14.5s
1 585667 87
2 198117 20
3 123360 6
4 117152 0
5 116181 0
6 116832 0
7 116421 0
8 116559 0
9 116344 0
14.5s
1 585667 87
2 198117 20
3 123360 6
4 117152 0
5 116181 0
6 116832 0
7 116421 0
8 116559 0
9 116344 0
14.3s
1 585667 87
2 198117 20
3 123360 6
4 117152 0
5 116181 0
6 116832 0
7 116421 0
8 116559 0
9 116344 0
14.4s
1 585667 87
2 198117 20
3 123360 6
4 117152 0
5 116181 0
6 116832 0
7 116421 0
8 116559 0
9 116344 0
15.5s
1 585667 87
2 198117 20
3 123360 6
4 117152 0
5 116181 0
6 116832 0
7 116421 0
8 116559 0
9 116344 0
15.9s
1 585667 87
2 198117 20
3 123360 6
4 117152 0
5 116181 0
6 116832 0
7 116421 0
8 116559 0
9 116344 0
14.9s
1 585667 87
2 198117 20
3 123360 6
4 117152 0
5 116181 0
6 116832 0
7 116421 0
8 116559 0
9 116344 0
15.4s
1 585667 87
2 198117 20
3 123360 6
4 117152 0
5 116181 0
6 116832 0
7 116421 0
8 116559 0
9 116344 0

Species count increment, with another for loop, with reduction

19.0s
1 585667 87
2 198117 20
3 123360 6
4 117152 0
5 116181 0
6 116832 0
7 116421 0
8 116559 0
9 116344 0
19.6s
1 585667 87
2 198117 20
3 123360 6
4 117152 0
5 116181 0
6 116832 0
7 116421 0
8 116559 0
9 116344 0
18.8s
1 585667 87
2 198117 20
3 123360 6
4 117152 0
5 116181 0
6 116832 0
7 116421 0
8 116559 0
9 116344 0
19.1s
1 585667 87
2 198117 20
3 123360 6
4 117152 0
5 116181 0
6 116832 0
7 116421 0
8 116559 0
9 116344 0
19.1s
1 585667 87
2 198117 20
3 123360 6
4 117152 0
5 116181 0
6 116832 0
7 116421 0
8 116559 0
9 116344 0
19.0s
1 585667 87
2 198117 20
3 123360 6
4 117152 0
5 116181 0
6 116832 0
7 116421 0
8 116559 0
9 116344 0
19.6s
1 585667 87
2 198117 20
3 123360 6
4 117152 0
5 116181 0
6 116832 0
7 116421 0
8 116559 0
9 116344 0
19.6s
1 585667 87
2 198117 20
3 123360 6
4 117152 0
5 116181 0
6 116832 0
7 116421 0
8 116559 0
9 116344 0
19.5s
1 585667 87
2 198117 20
3 123360 6
4 117152 0
5 116181 0
6 116832 0
7 116421 0
8 116559 0
9 116344 0