search

google / nccl-fastsocket

NCCL Fast Socket is a transport layer plugin to improve NCCL collective communication performance on Google Cloud.
Other
109 stars 13 forks source link

what's the key actions that fastsocket do to improve NCCL collective communication performance? #7

Open yangxingwu opened 1 year ago

yangxingwu commented 1 year ago

as README stated

NCCL Fast Socket is based on TCP/IP communication and uses a number of techniques to achieve better and more consistent performance, especially with 100 Gbps networking on Google Cloud

  1. does this applied to a self-hosted clusters with 100Gbps Nics?
  2. what's the key actions that fastsocket do to improve NCCL collective communication performance?
ixp9891 commented 1 year ago

mark