.png){kind=icon}
.png){kind=icon}
Return
Understanding node connection modes in Multi-Rail Fat-tree
.png){kind=icon}
PRE
AI
DOI：10.1016/j.jpdc.2022.04.019.png)
.png){kind=icon}
Original request for help.png){kind=icon}
Share.png){kind=icon}
Save
Abstract
En 中文
Using Multi-Rail networks has become a popular choice for many leading HPC systems to overcome bandwidth limitations. Little is known, theoretically or practically, how the connection modes between multi-port nodes and switching network influence the performance of system. This work provides a detailed analysis of different node connection modes in Multi-Rail Fat-tree, which involves Single-Plane Multi-Rail Fat-tree and Multi-Plane Multi-Rail Fat-tree. We evaluate different node connection modes by theoretical analysis and flit level simulation. We show that there are great differences across various node connection modes. Among the key differences are the following: cost, average shortest path length, fault-tolerance, path diversity, latency and throughput. In addition, we propose MR-tree and MP-tree respectively to gain a deep understanding of relevant issues and improve the network performance. Our findings leave open the possibility that optimization of node connection mode can yield better results for Multi-Rail Fat-tree. (C) 2022 Elsevier Inc. All rights reserved.
Keywords:
HPC
Topology
Fat-tree
Multi-Rail
Connection mode
Journal
IF:
4
Papers:
3.8K
Citations:
4.8K