NFS

From BGL

Jump to: navigation, search

Some notes from the BG/L administrator course and experience at EPCC

  • Any scripts on the IO nodes to mount filesystems must return before ciod starts on the node. Once ciod is up, the jobs can start.
  • IBM Tuning found better performance on NFS using TCP rather than UDP, NFS3, Async. R/W size of 32K,
  • We have tuned the our RedHat NFS servers to bluegene as follows:

/etc/sysctl.conf 

# Tunining tweaks for NFS
vm.max-readahead=256
vm.min-readahead=128
net.core.rmem_default=262144
net.core.rmem_max=262144
net.core.wmem_default=262144
net.core.wmem_max=262144
net.core.netdev_max_backlog=10000
net.ipv4.neigh.default.gc_thresh1=16384
net.ipv4.neigh.default.gc_thresh2=16384
net.ipv4.neigh.default.gc_thresh3=16384

and /etc/sysconfig/nfs: 

#number of nfsd daemons
RPCNFSDCOUNT=32
Retrieved from "http://wiki.bgl.mcs.anl.gov/wiki/NFS"
Personal tools