search

jasperzhong / read-papers-and-code

My paper/code reading notes in Chinese
43 stars 3 forks source link

NSDI '23 | BGL: GPU-Efficient GNN Training by Optimizing Graph Data I/O and Preprocessing #379

Closed jasperzhong closed 8 months ago

jasperzhong commented 8 months ago

https://www.usenix.org/system/files/nsdi23-liu-tianfeng.pdf

jasperzhong commented 8 months ago

之前没怎么看graph partitioning那个部分.

大致意思就是将一个图分成多个block（coarsening），然后用一个heuristics assign block到各个partitions，最后再uncoarsening一下，得到最后的partition. 这其实是比较traditional的multi-level graph partitioning.