[netperf-dev] short circuit for confidence intervals?

[Rick Jones]

Folks -

While running some scripts I've noticed that on some tests, TCP_RR 
especially, I often notice that the confidence intervals for the trio of 
result, local CPU and remote CPU aren't hit, overall but the confidence 
interval for the result is often spot-on within < 5 iterations.

I'm wondering if it would be worthwhile to provide some sort of "I only 
_really_ care about confidence on the result" switch whereby netperf 
(netperf2) still calculates confidence for all three, but will quit when 
confidence for the result is hit regardless of confidence on CPU utils.

I'm thinking of a new global switch which would be examined by the 
confidence code in netlib.c and calculate "confidence" based on 
results_confid only.  There would remain the issue of the test banners 
asserting +/- N% which would have to be ammended when this switch was 
enabled, and then the prospect of some additional -v 2 output to give 
the intervals for the CPU utils.

Or should I just leave things as they are and cope with the longer 
script runtimes when it runs all the way out to 30 iterations ?-)

rick jones