[net-perf] bwctl testing

[Michael Van Norman]

Michael Van Norman wrote:
> A bug in the interaction 
> between iperf and bwctl loses the results when originating some test 
> from and I need to track it down and fix it.  In case anybody else has 
> seen it and has a solution, the temporary file that iperf writes its 
> results to is being truncated (i.e. the full results don't actually make 
> it into the file).

This ended up being a performance problem with iperf.  Seemed to have 
crept in somewhere on the way to kernel 2.6.11.  A bit of tuning for the 
new kernel seems to have fixed things up.

For the curious:

--- snip ---
#
# /etc/sysctl.conf - Configuration file for setting system variables
#

# increase TCP max buffer size
net.core.rmem_max=16777216
net.core.wmem_max=16777216
# increase Linux autotuning TCP buffer limits
# min, default, and max number of bytes to use
net.ipv4.tcp_rmem="4096 87380 16777216"
net.ipv4.tcp_wmem="4096 65536 16777216"

# don't cache ssthresh from previous connection
net.ipv4.tcp_no_metrics_save=1
# recommended to increase this for 1000 BT or higher
net.core.netdev_max_backlog=2500
--- snip ---

That and using ethtool to turn off tcp segmentation offload (e.g. 
"ethtool -K gig0 tso off").

Culled from http://www-didc.lbl.gov/TCP-tuning/linux.html

/Mike