[netperf-talk] Contemplating UI changes involving HISTOGRAM support

Rick Jones rick.jones2 at hp.com
Wed Feb 7 18:06:31 PST 2007


Folks -

Those of you watching netperf-dev and/or following the top-of-trunk at 
http://www.netperf.org/svn/netperf2/trunk/ may have noticed that there 
have been some recent changes involving --enable-histogram.

Previously, with --enable-histogram, the histogram itself would only be 
displayed if the verbosity level was 2 or greater yet timestamping would 
happen all the time.  This meant a measurable effect on results even if 
a histogram wasn't going to be displayed.

I've now started bracketing the timestamping with "if (verbosity > 1)" 
calls - so the overhead of --enable-histogram is negligible if verbosity 
is default (1) and only kicks-in if one would be displaying the histogram.

This then begs the question - should --enable-histogram become the 
default for netperf 2.4.3?

There is other data displayed at -v 2 - data displayed even if 
--enable-histogram isn't enabled.  If --enable-histogram were to become 
the default, then -v 2 would start to have measureable overhead it 
didn't have before.

I _could_ address that by shifting histogram timestamping and output to 
verbosity > 2 (ie 3 or more) but that would then be a change from 
previous behaviour too.

Thoughts?

happy benchmarking,

rick jones


More information about the netperf-talk mailing list