Skip to content

Commit

Permalink
Merge branch 'eurosys22-final-version-sxn-dev' into 'eurosys22-final-…
Browse files Browse the repository at this point in the history 
…version'

Add gpu to pinsage with dgl in table4

See merge request gnn/samgraph!92
  • Loading branch information
@dynamicheart
dynamicheart committed Mar 11, 2022
2 parents 1ca780f + c6d1ca6 commit de73769
Show file tree
Hide file tree
Showing 5 changed files with 28 additions and 25 deletions.
2 changes: 1 addition & 1 deletion README.md
View file
Original file line number Diff line number Diff line change
Expand Up @@ -220,4 +220,4 @@ FGNN is released under the [Apache License 2.0](http://www.apache.org/licenses/L
## Academic and Conference Papers
[**EuroSys**] FGNN: A Factored System for Sample-based GNN Training over GPUs. Jianbang Yang, Dahai Tang, Xiaoniu Song, Lei Wang, Qiang Yin, Rong Chen, Wenyuan Yu, Jingren Zhou. Proceedings of the 17th European Conference on Computer Systems, Rennes, France, April, 2022.
[**EuroSys**] GNNLab: A Factored System for Sample-based GNN Training over GPUs. Jianbang Yang, Dahai Tang, Xiaoniu Song, Lei Wang, Qiang Yin, Rong Chen, Wenyuan Yu, Jingren Zhou. Proceedings of the 17th European Conference on Computer Systems, Rennes, France, April, 2022.
32 changes: 16 additions & 16 deletions exp/table4/README.md
View file
Original file line number Diff line number Diff line change
Expand Up @@ -11,10 +11,10 @@ The goal of this experiment is to show the overall end2end performance of DGL, P

- Paper's configurations: **8x16GB** NVIDIA V100 GPUs, **2x24** cores Intel 8163 CPU
- For other hardware configurations, you may need to modify the ①Number of GPU ②Number of CPU threads. ③Number of vertex(in percentage, 0<=pct. <=1) to be cached.
- **DGL:** Modify `L69(#GPU), L110(#GPU)` in `run.py`.
- **PyG:** Modify `L142(#CPU threads), L145(#GPU)` in `run.py`.
- **SGNN:** Modify `L184(#CPU threads), L203(#GPU), L204-L254(#Cache Percentage)` in `run.py`.
- **FGNN:** Modify `L290(#CPU threads), L304-L351(#CPU, #Cache percentage)` in `run.py`.
- **DGL:** Modify `L69(#GPU), L113(#GPU)` in `run.py`.
- **PyG:** Modify `L145(#CPU threads), L148(#GPU)` in `run.py`.
- **SGNN:** Modify `L187(#CPU threads), L206(#GPU), L207-L254(#Cache Percentage)` in `run.py`.
- **FGNN:** Modify `L293(#CPU threads), L310-L354(#CPU, #Cache percentage)` in `run.py`.



Expand Down Expand Up @@ -76,18 +76,18 @@ output_2022-01-29_18-17-21
```sh
> cat output_2022-01-29_18-17-21/table4.dat
GNN Models Dataset DGL PyG SGNN FGNN
GCN PR 1.18 11.28 0.23 0.33(3S)
GCN TW 3.40 11.48 1.35 0.41(2S)
GCN PA 4.07 14.58 2.75 0.86(2S)
GCN UK X 13.30 X 1.20(2S)
GraphSAGE PR 0.71 7.45 0.07 0.11(4S)
GraphSAGE TW 1.63 7.50 0.32 0.17(2S)
GraphSAGE PA 2.17 8.64 0.95 0.28(2S)
GraphSAGE UK X 8.95 1.59 0.51(1S)
PinSAGE PR 0.80 X 0.31 0.40(1S)
PinSAGE TW 2.01 X 0.81 0.51(1S)
PinSAGE PA 2.56 X 1.72 1.03(1S)
PinSAGE UK X X X 1.43(1S)
GCN PR 1.23 11.28 0.23 0.33(3S)
GCN TW 3.40 11.48 1.42 0.43(2S)
GCN PA 4.14 14.58 2.68 0.81(2S)
GCN UK X 13.30 X 1.29(2S)
GraphSAGE PR 0.71 7.45 0.11 0.13(4S)
GraphSAGE TW 1.61 7.50 0.33 0.17(2S)
GraphSAGE PA 2.34 8.64 1.00 0.28(2S)
GraphSAGE UK X 8.95 1.65 0.55(1S)
PinSAGE PR 0.86 X 0.33 0.41(1S)
PinSAGE TW 2.10 X 0.89 0.54(1S)
PinSAGE PA 2.59 X 1.69 1.03(1S)
PinSAGE UK X X X 1.53(1S)
```
Expand Down
2 changes: 1 addition & 1 deletion exp/table4/logtable_def.py
View file
Original file line number Diff line number Diff line change
Expand Up @@ -60,7 +60,7 @@ def get_dgl_pinsage_logtable():
num_col=1
).update_col_definition(
col_id=0,
definition='epoch_time:total'
definition='epoch_time'
).update_row_definition(
row_id=0,
col_range=[0, 0],
Expand Down
11 changes: 7 additions & 4 deletions exp/table4/run.py
View file
Original file line number Diff line number Diff line change
Expand Up @@ -16,7 +16,7 @@

DGL_APP_DIR = os.path.join(HERE, '../../example/dgl/multi_gpu')
# Simulate PinSAGE(GPU version) on DGL
DGL_PINSAGE_APP_DIR = os.path.join(HERE, '../../example/samgraph/sgnn_dgl')
DGL_PINSAGE_APP_DIR = os.path.join(HERE, '../../example/dgl/multi_gpu')
PYG_APP_DIR = os.path.join(HERE, '../../example/pyg/multi_gpu')
SGNN_APP_DIR = os.path.join(HERE, '../../example/samgraph/sgnn')
FGNN_APP_DIR = os.path.join(HERE, '../../example/samgraph/multi_gpu')
Expand Down Expand Up @@ -102,12 +102,15 @@ def dgl_pinsage_overall_test():
).override(
'num_epoch',
[NUM_EPOCH]
).override(
'BOOL_use_gpu_sampling',
['use_gpu_sampling']
).override(
'BOOL_pipeline',
['pipeline']
['pipelining']
).override(
'num_worker',
[8],
'devices',
['0 1 2 3 4 5 6 7'],
# ).override(
# 'BOOL_validate_configs',
# ['validate_configs']
Expand Down
6 changes: 3 additions & 3 deletions exp/table5/README.md
View file
Original file line number Diff line number Diff line change
Expand Up @@ -81,17 +81,17 @@ output_2022-01-29_18-54-46
GCN | PR | 0.39 3.22 1.30 | 0.32 = 0.31 + 0.01 0.05 (1.00%, 1.00%) 1.30 | 0.39 = 0.30 + 0.01 + 0.08 0.15 (1.00%, 1.00%) 1.30
GCN | TW | 0.90 11.04 1.64 | 0.31 = 0.28 + 0.03 3.51 (0.01%, 0.32%) 1.70 | 0.36 = 0.26 + 0.03 + 0.07 0.81 (0.25%, 0.89%) 1.73
GCN | PA | 1.67 12.73 4.21 | 0.84 = 0.73 + 0.11 0.32 (0.07%, 0.98%) 4.48 | 0.95 = 0.68 + 0.11 + 0.17 0.50 (0.21%, 0.99%) 4.46
GCN | UK | 1.49 18.46 3.44 | 0.42 = 0.42 + 0.00 8.45 (0.00%, 0.00%) 3.51 | 0.55 = 0.38 + 0.03 + 0.13 3.06 (0.14%, 0.70%) 3.57
GCN | UK | X X X | X = X + X X ( X%, X%) X | 0.55 = 0.38 + 0.03 + 0.13 3.06 (0.14%, 0.70%) 3.57
--------------------------------------------------------------------------------------------------------------------------------------------------------------------
GSG | PR | 0.15 2.47 0.24 | 0.16 = 0.15 + 0.01 0.03 (1.00%, 1.00%) 0.28 | 0.20 = 0.15 + 0.01 + 0.04 0.08 (1.00%, 1.00%) 0.29
GSG | TW | 0.43 5.61 0.49 | 0.13 = 0.11 + 0.02 0.63 (0.15%, 0.82%) 0.54 | 0.15 = 0.11 + 0.01 + 0.03 0.48 (0.32%, 0.89%) 0.52
GSG | PA | 0.64 7.07 1.40 | 0.39 = 0.33 + 0.06 0.19 (0.11%, 0.98%) 1.50 | 0.45 = 0.31 + 0.06 + 0.08 0.35 (0.25%, 0.99%) 1.47
GSG | UK | 0.61 10.10 1.26 | 0.19 = 0.19 + 0.00 4.31 (0.00%, 0.00%) 1.26 | 0.26 = 0.18 + 0.02 + 0.06 1.35 (0.18%, 0.72%) 1.27
GSG | UK | X X X | 0.19 = 0.19 + 0.00 4.31 (0.00%, 0.00%) 1.26 | 0.26 = 0.18 + 0.02 + 0.06 1.35 (0.18%, 0.72%) 1.27
--------------------------------------------------------------------------------------------------------------------------------------------------------------------
PSG | PR | 0.48 1.97 1.91 | 0.18 = 0.17 + 0.01 0.03 (1.00%, 1.00%) 1.97 | 0.20 = 0.15 + 0.01 + 0.04 0.07 (1.00%, 1.00%) 1.93
PSG | TW | 0.95 6.16 2.87 | 0.25 = 0.23 + 0.02 1.19 (0.04%, 0.62%) 2.94 | 0.28 = 0.21 + 0.02 + 0.05 0.58 (0.26%, 0.86%) 2.86
PSG | PA | 2.46 5.95 6.30 | 0.57 = 0.52 + 0.05 0.22 (0.06%, 0.97%) 6.80 | 0.62 = 0.48 + 0.05 + 0.09 0.34 (0.22%, 0.97%) 6.94
PSG | UK | 14.98 15.71 7.56 | 0.49 = 0.49 + 0.00 6.81 (0.00%, 0.00%) 7.71 | 0.63 = 0.47 + 0.03 + 0.13 3.39 (0.13%, 0.57%) 8.00
PSG | UK | X X X | X = X + X X ( X%, X%) X | 0.63 = 0.47 + 0.03 + 0.13 3.39 (0.13%, 0.57%) 8.00
--------------------------------------------------------------------------------------------------------------------------------------------------------------------
```
Expand Down

0 comments on commit de73769

Please sign in to comment.