{"id":13605298,"url":"https://github.com/MachineLearningSystem/MGG-OSDI23-AE","last_synced_at":"2025-04-12T05:32:19.599Z","repository":{"id":185461850,"uuid":"632229734","full_name":"MachineLearningSystem/MGG-OSDI23-AE","owner":"MachineLearningSystem","description":"OSDI'23-MGG-Artifact","archived":false,"fork":true,"pushed_at":"2023-04-25T01:00:29.000Z","size":1086,"stargazers_count":0,"open_issues_count":0,"forks_count":0,"subscribers_count":0,"default_branch":"master","last_synced_at":"2024-11-07T10:40:56.863Z","etag":null,"topics":[],"latest_commit_sha":null,"homepage":"","language":null,"has_issues":false,"has_wiki":null,"has_pages":null,"mirror_url":null,"source_name":"YukeWang96/MGG_OSDI23","license":null,"status":null,"scm":"git","pull_requests_enabled":true,"icon_url":"https://github.com/MachineLearningSystem.png","metadata":{"files":{"readme":"README.md","changelog":null,"contributing":null,"funding":null,"license":null,"code_of_conduct":null,"threat_model":null,"audit":null,"citation":null,"codeowners":null,"security":null,"support":null,"governance":null}},"created_at":"2023-04-25T01:36:28.000Z","updated_at":"2023-04-25T01:36:28.000Z","dependencies_parsed_at":null,"dependency_job_id":"c399c2ad-0725-4ce6-ad1f-43ebc6c2c5b8","html_url":"https://github.com/MachineLearningSystem/MGG-OSDI23-AE","commit_stats":null,"previous_names":["machinelearningsystem/mgg-osdi23-ae"],"tags_count":0,"template":false,"template_full_name":null,"repository_url":"https://repos.ecosyste.ms/api/v1/hosts/GitHub/repositories/MachineLearningSystem%2FMGG-OSDI23-AE","tags_url":"https://repos.ecosyste.ms/api/v1/hosts/GitHub/repositories/MachineLearningSystem%2FMGG-OSDI23-AE/tags","releases_url":"https://repos.ecosyste.ms/api/v1/hosts/GitHub/repositories/MachineLearningSystem%2FMGG-OSDI23-AE/releases","manifests_url":"https://repos.ecosyste.ms/api/v1/hosts/GitHub/repositories/MachineLearningSystem%2FMGG-OSDI23-AE/manifests","owner_url":"https://repos.ecosyste.ms/api/v1/hosts/GitHub/owners/MachineLearningSystem","download_url":"https://codeload.github.com/MachineLearningSystem/MGG-OSDI23-AE/tar.gz/refs/heads/master","host":{"name":"GitHub","url":"https://github.com","kind":"github","repositories_count":248524053,"owners_count":21118607,"icon_url":"https://github.com/github.png","version":null,"created_at":"2022-05-30T11:31:42.601Z","updated_at":"2022-07-04T15:15:14.044Z","host_url":"https://repos.ecosyste.ms/api/v1/hosts/GitHub","repositories_url":"https://repos.ecosyste.ms/api/v1/hosts/GitHub/repositories","repository_names_url":"https://repos.ecosyste.ms/api/v1/hosts/GitHub/repository_names","owners_url":"https://repos.ecosyste.ms/api/v1/hosts/GitHub/owners"}},"keywords":[],"created_at":"2024-08-01T19:00:57.156Z","updated_at":"2025-04-12T05:32:19.326Z","avatar_url":"https://github.com/MachineLearningSystem.png","language":null,"readme":"# Artifact for OSDI'23 paper \n\u003e Yuke Wang, et al. *Accelerating Graph Neural Networks with Fine-grained intra-kernel Communication-Computation Pipelining on Multi-GPU Platforms.* OSDI'23.\n\n[![DOI](https://zenodo.org/badge/DOI/10.5281/zenodo.7853910.svg)](https://doi.org/10.5281/zenodo.7853910)\n\n# 1. Setup (Skip to Section-2 if evaluated on provided GCP)\n\n## 1.1. Clone this project from Github.\n```\ngit clone --recursive git@github.com:YukeWang96/MGG-OSDI23-AE.git\n```\n\n## 1.2. Download libraries and datasets.\n+ Download libraries (`cudnn-v8.2, nvshmem_src_2.0.3-0, openmpi-4.1.1`).\n```\nwget https://storage.googleapis.com/mgg_data/local.tar.gz\ntar -zxvf local.tar.gz \u0026\u0026 rm local.tar.gz\ntar -zxvf local/nvshmem_src_2.0.3-0/build_cu112.tar.gz\n```\n+ Setup baseline DGL\n```\ncd dgl_pydirect_internal\nwget https://storage.googleapis.com/mgg_data/graphdata.tar.gz \u0026\u0026 tar -zxvf graphdata.tar.gz \u0026\u0026 rm graphdata.tar.gz\ncd ..\n```\n\n+ Setup baseline ROC\n```\ncd roc-new\ngit submodule update --init --recursive\nwget https://storage.googleapis.com/mgg_data/data.tar.gz \u0026\u0026 tar -zxvf data.tar.gz \u0026\u0026 rm -rf data.tar.gz\n```\nor \n```\ngsutil cp -r gs://mgg_data/roc-new/ .\n```\n\n## 1.3. Launch Docker for MGG.\n```\ncd Docker \n./launch.sh\n```\n\n## 1.4. Compile implementation.\n```\nmkdir build \u0026\u0026 cd build \u0026\u0026 cmake .. \u0026\u0026 cd ..\n./build.sh\n```\n# 2. Run initial test experiment.\n+ Please try study experiments in below **Section-3.4** and **Section-3.5**\n\n\n# 3. Reproduce the major results from paper.\n\n## 3.1 Compare with UVM on 4xA100 and 8xA100 (Fig.8a and Fig.8b).\n```\n./0_run_MGG_UVM_4GPU_GCN.sh\n./0_run_MGG_UVM_4GPU_GIN.sh\n./0_run_MGG_UVM_8GPU_GCN.sh\n./0_run_MGG_UVM_8GPU_GIN.sh\n```\n\u003e Note that the results can be found at `Fig_8_UVM_MGG_4GPU_GCN.csv`, `Fig_8_UVM_MGG_4GPU_GIN.csv`, `Fig_8_UVM_MGG_8GPU_GCN.csv`, and `Fig_8_UVM_MGG_8GPU_GIN.csv`.\n\n\n## 3.2 Compare with DGL on 8xA100 for GCN and GIN (Fig.7a and Fig.7b).\n```\n./launch_docker.sh\ncd gcn/\n./0_run_gcn.sh\ncd ../gin/\n./0_run_gin.sh\n```\n\n\u003e Note that the results can be found at `1_dgl_gin.csv` and `1_dgl_gcn.csv` and our MGG reference is in `MGG_GCN_8GPU.csv` and `MGG_8GPU_GIN.csv`.\n\n\n## 3.3 Compare with ROC on 8xA100 (Fig.9).\n```\ncd roc-new/docker\n./launch.sh\n```\n\u003e Note that the results can be found at `Fig_9_ROC_MGG_8GPU_GCN.csv` and `Fig_9_ROC_MGG_8GPU_GIN.csv`.\n\nResults of ROC is similar as\n\n| Dataset       | Time (ms) |\n|---------------|----------:|\n| reddit        |    425.67 |\n| enwiki-2013   |    619.33 |\n| it-2004       |   5160.18 |\n| paper100M     |   8179.35 |\n| ogbn-products |    529.74 |\n| ogbn-proteins |    423.82 |\n| com-orkut     |    571.62 |\n\n\n## 3.4 Compare NP with w/o NP (Fig.10a).\n```\npython 2_MGG_NP.py\n```\n\u003e Note that the results can be found at `MGG_NP_study.csv`. Similar to following table.\n\n| Dataset      | MGG_WO_NP | MGG_W_NP | Speedup (x) |\n|--------------|----------:|---------:|------------:|\n| Reddit       |    76.797 |   16.716 |       4.594 |\n| enwiki-2013  |   290.169 |   88.249 |       3.288 |\n| ogbn-product |    86.362 |   26.008 |       3.321 |\n\n## 3.5 Compare WL with w/o WL (Fig.10b).\n\n```\npython 3_MGG_WL.py\n```\n\u003e Note that the results can be found at `MGG_WL_study.csv`. Results are similar to\n\n| Dataset      | MGG_WO_NP | MGG_W_NP | Speedup (x) |\n|--------------|----------:|---------:|------------:|\n| Reddit       |    75.035 |    18.92 |       3.966 |\n| enwiki-2013  |   292.022 |  104.878 |       2.784 |\n| ogbn-product |    86.632 |   29.941 |       2.893 |\n\n## 3.6 Compare API (Fig.10c).\n```\npython 4_MGG_API.py\n```\n\u003e Note that the results can be found at `MGG_API_study.csv`. Results are similar to \n\n| Norm.Time w.r.t. Thread | MGG_Thread | MGG_Warp | MGG_Block |\n|-------------------------|------------|----------|-----------|\n| Reddit                  | 1.0        | 0.299    | 0.295     |\n| enwiki-2013             | 1.0        | 0.267    | 0.263     |\n| ogbn-product            | 1.0        | 0.310    | 0.317     |\n\n\n## 3.7 Design Space Search (Fig.11a)\n```\npython 5_MGG_DSE_4GPU.py\n```\n\u003e Note that the results can be found at `Reddit_4xA100_dist_ps.csv` and `Reddit_4xA100_dist_wpb.csv`. Results similar to \n\n+ `Reddit_4xA100_dist_ps.csv`\n\n| dist\\ps |      1 |      2 |      4 |      8 |     16 |     32 |\n|---------|-------:|-------:|-------:|-------:|-------:|-------:|\n| 1       | 17.866 | 17.459 | 16.821 | 16.244 | 16.711 | 17.125 |\n| 2       | 17.247 | 16.722 | 16.437 | 16.682 | 17.053 | 17.808 |\n| 4       | 16.826 |  16.41 | 16.583 | 17.217 | 17.627 | 18.298 |\n| 8       | 16.271 | 16.725 | 17.193 | 17.655 | 18.426 |  18.99 |\n| 16      | 16.593 | 17.214 | 17.617 | 18.266 | 19.009 | 19.909 |\n\n+ `Reddit_4xA100_dist_wpb.csv`\n\n| dist\\wpb |      1 |      2 |      4 |      8 |     16 |\n|----------|-------:|-------:|-------:|-------:|-------:|\n| 1        | 34.773 | 23.164 | 16.576 | 15.235 | 16.519 |\n| 2        | 34.599 | 23.557 | 17.254 | 15.981 |  19.56 |\n| 4        | 34.835 | 23.616 | 17.674 | 17.034 | 22.084 |\n| 8        | 34.729 | 23.817 | 18.302 | 18.708 | 25.656 |\n| 16       | 34.803 | 24.161 | 18.879 |  23.44 | 32.978 |\n\n\n\n```\npython 5_MGG_DSE_8GPU.py\n```\n\u003e Note that the results can be found at `Reddit_8xA100_dist_ps.csv` and `Reddit_8xA100_dist_wpb.csv`.\n\n\n\n## Reference\n* **NVIDIA OpenSHMEM Library (NVSHMEM) Documentation.** \u003cbr\u003e\nhttps://docs.nvidia.com/nvshmem/api/index.html\n\n* **NVIDIA Unified Memory.** \u003cbr\u003e\nhttps://developer.nvidia.com/blog/unified-memory-cuda-beginners/\n\n* **cuDNN Example for MNIST.** \u003cbr\u003e\nhttps://github.com/haanjack/mnist-cudnn\n\n* [**Deep Graph Library**](https://github.com/dmlc/dgl) \u003cbr\u003e\nWang, Minjie, et al. \n**Deep graph library: A graph-centric, highly-performant package for graph neural networks.**. *The International Conference on Learning Representations (ICLR'19).*\n\n* [**ROC**](https://github.com/jiazhihao/ROC) \u003cbr\u003e\nJia, Zhihao, et al. \n**Improving the accuracy, scalability, and performance of graph neural networks with roc.** *Proceedings of Machine Learning and Systems 2 (MLsys'20).*\n\n* [**GNNAdvisor**](https://github.com/YukeWang96/OSDI21_AE) \u003cbr\u003e\nWang, Yuke, et al. **GNNAdvisor: An adaptive and efficient runtime system for GNN acceleration on GPUs.** *15th USENIX symposium on operating systems design and implementation (OSDI'21)*.\n\n* [**GE-SpMM**](https://github.com/hgyhungry/ge-spmm) \u003cbr\u003e\nHuang, Guyue, et al. **Ge-spmm: General-purpose sparse matrix-matrix multiplication on gpus for graph neural networks.** *International Conference for High Performance Computing, Networking, Storage and Analysis (SC'20)*.\n","funding_links":[],"categories":["Paper-Code"],"sub_categories":["GNN"],"project_url":"https://awesome.ecosyste.ms/api/v1/projects/github.com%2FMachineLearningSystem%2FMGG-OSDI23-AE","html_url":"https://awesome.ecosyste.ms/projects/github.com%2FMachineLearningSystem%2FMGG-OSDI23-AE","lists_url":"https://awesome.ecosyste.ms/api/v1/projects/github.com%2FMachineLearningSystem%2FMGG-OSDI23-AE/lists"}