基金Ye YUAN is supported by the National Key R&D Program of China(Grant No.2022YFB2702100)the National Natural Science Foundation of China(NSFC)(Grant Nos.61932004,62225203,U21A20516)+5 种基金Hangxu JI is supported by the NSFC(Grant No.62402096)the Guangdong Basic and Applied Basic Research Foundation(Grant No.2023A1515110268)Yishu WANG is supported by the NSFC(Grant No.62302084)Yuliang MA is supported by the NSFC(Grant No.62002054)the Fundamental Research Funds for the Central Universities of China(Grant No.N2304014)the Liaoning Provincial Natural Science Foundation Joint Fund(Grant No.2023-MSBA-080).
文摘1Introduction Graph partitioning is essential for large-scale distributed graph processing,as partitioning strategies directly affect graph algorithm performance[1].To ensure reliability and scalability,many graphbased applications[2](e.g.,Facebook,Weibo)deploy their services over geo-distributed datacenters(DCs),posing challenges for existing partition methods.These methods may struggle with heterogeneity(e.g.,network bandwidth)or overlook structural properties(e.g.,community structure)during optimization.
