1、AliGraph:?赵 昆?Analysis of Large-scale GNN System Architecture Based on AliGraphContents01目录AliGraph?02?AliGraph平台简介AliGraph Overview01PAI：阿里巴巴机器学习平台Platform of Artificial Intelligence in AlibabaPAI机器学习平台特性Features of PAI?https:/ of GNNGNNGNN低=向量空间低=向量空间G=(V,E G=(V,E V V=V)rt)x E=E(g)=V)rt)x E=E(g)N

2、N)ura,)ura,N)tworkN)twork实践中的挑战Challenges in Practice数据不规则稀疏性，非结构化，属性同构，异构，有向，无向DataCleaner规模巨大千亿边、百亿点样本指数级膨胀?TB?CPU?计算难扩展图算子种类变化多无法结合深度学习?用AD望：一体化G决方EUsers Expect Solutions from Data to Model for GNNA APPPPGNN AlgorithmsGraph&TensorGraph&TensorEngineGraph DataGraph DataLarge-s-ale&omogeneous&eterog

3、eneous Attri,utedAliGraph：云上的“图+深度学+”平台A GNN Platform on Alibaba Cloud?Industry Level?PB?End to End?IDE?For Research and Production系统架构与核心技术System Architecture and Key Technologies02GNN编程范式GNN Programming Paradigmhop=1hop=21.Sampling2.Vectorization3.Aggregate4.CombineVertex IdVertex AttributesEdge A

4、ttributesEdge IdAggregate outputVertex outputV1V2E1E2V0一体化系统实现An Integrated System Implementation for GNNServer 0Server Server 2Server nTensor EngineGraph EngineData BridgePythonTensorFlow&PyTorchGraph QuerySamplerAggregatorCombinerLarge-scaleHeterogeneousAttributedGNNRPCFriendlyFlexibleRobustEffect

5、ive简洁的用户接4Laconic User Interface点3相2g=Graph()g.add_nodes(“node_source”,“node_format”).add_edges(“edge_source”,“edge_format”).init().2.3.4.Graph B1ildi.gg.V(“user”).shuffle().batch(512).outE(“user-click-item”).sample(10).by(“edge_weight”).run().2.3.4.Graph Sampli.g高效的图构建过程Effective Graph Building SER

6、VER iData QueueData QueueData QueueLocalPROCESSINGTHREADPROCESSINGTHREADPROCESSINGTHREADData QueueData QueueData QueuePROCESSINGTHREADPROCESSINGTHREADPROCESSINGTHREADLOADING THREADRPCRPCLocalLOADING THREADLOADING THREADRPCRPCSERVERjLOADING THREADBatch Reading Batch Reading PartitioningPartitioningPa

7、rtitioningPartitioningBatch Reading Batch Reading Graph ParsingGraph ParsingGraph ParsingGraph ParsingGraph ParsingGraph ParsingUpdate StorageUpdate StorageUpdate StorageUpdate StorageUpdate StorageUpdate Storage多跳采SG速R制Cache to Empower Multi-Hop Graph SamplingCa0hing 2 hopk-1 hopk hopS2rv2r 0S2rv2r

8、 1S2rv2r n多跳缓存P速一倍 图顶TU度数V合)oA2r-laA分布 LC顶TU重要性)2r02ntag2 of Ca0h21 V2rti02sTim2 Cost(ms AliGraphAliGraphRan1om Ca0h2Ran1om Ca0h2(R-Ca0h2(R-Ca0h2可扩展的算子类型Flexible and Pluggable OperatorsUser nterfaceUser nterfaceDistributeDistributeRuntimeRuntimeDistributeDistribute StorageStorageProgrammingProgrammi

9、ng nterfacenterfaceProgramming Programming nterfacenterfaceOperatorOperatorOperatorOperatorOperatorOperatorOperatorOperator系n数USystem Performance?GNNi法P发周期pe80%?模型训m时长pe%0%?N支T的Gr/ph Em0211583i法10余种每天节d存储300TB mo12l每天节diB万CPU时 mo12lb亿边lG构建2分钟多跳批量G采样C毫gl内置G采样i法10余种E扩展的i子种类用户raoh已落地的应用Applications新零售人际关系淘宝推荐搜索反作弊线上支付淘宝搜索网络安全系统的发展方向Evolution Direction of the System生态多样性生态多样性系统延展性系统延展性GNNGNN算法驱动算法驱动THANKSTHANKS!THANKS!