[Beowulf] substantial RX packet drops during Pallas over e1000 (Rocks 4.1)
hahn at physics.mcmaster.ca
Wed May 17 14:53:37 PDT 2006
> Running Rocks 4.1 on a 30 node system and seeing serious RX packet
> loss, drops and overruns while running heavy MPI i/o over e1000. I have
e1000, at least some chips, have interrupt-mitigation.
you should use ethtool to query these settings.
you might also tweak vm.min_free_kbytes. how about
the qlen as seen by ip/ifconfig? also, have you tuned
the *mem settings for the net stack itself like net.core.rmem_max?
More information about the Beowulf