图计算——我们能否兼得 OLAP 和 OLTP 的优势？.pdf

1、 2025,Amazon Web Services,Inc.or its affiliates.All rights reserved.2025,Amazon Web Services,Inc.or its affiliates.All rights reserved.2025,Amazon Web Services,Inc.or its affiliates.All rights reserved.INV502mit V.atalyrek(he/him)Amazon Scholar,Amazon Neptune,AWSProfessor,Georgia Institute of Techno

2、logyGraph Computations:Can we have best of two worlds:OLAP and OLTP?Yes!Well,almost.2025,Amazon Web Services,Inc.or its affiliates.All rights reserved.Landscape of current“Graph World”Enterprise Graph FrameworksHPC Graph AnalyticsGraph Databases3High performance for specific graph problemsGeneralize

3、d graph query languages with consistency and durabilityGeneralized APIs for graph processingOLTPOLAP3:30pm Today-DAT447-Using graphs over your data lake to power generative AI Applications 2025,Amazon Web Services,Inc.or its affiliates.All rights reserved.C S 1 0 1 H P C G R A P H A N A L Y T I C SG

4、raph is a Sparse MatrixCompressed Sparse Row/Column(CSR/CSC)Coordinate Format(COO)Graph Storage Models 2025,Amazon Web Services,Inc.or its affiliates.All rights reserved.Adjacency matrix For vertices Square matrix:Undirected-binary,symmetricGraph is a matrix136425001100001100110001110010000100001000

5、112233445566 2025,Amazon Web Services,Inc.or its affiliates.All rights reserved.Adjacency matrix For vertices Square matrix:Undirected-binary,symmetric Directed-binary,not symmetricGraph is a matrix136425001100001000010001010000000100000000 2025,Amazon Web Services,Inc.or its affiliates.All rights r

6、eserved.Adjacency matrix For vertices Square matrix:Undirected-binary,symmetric Directed-binary,not symmetric Weighted-not binary How to store matrices in memory?Graph is a matrix1364250010150000300000300002002500000005000000000101530502520 2025,Amazon Web Services,Inc.or its affiliates.All rights r