Fix bug in percentage increase computation for cache sensitivity analysis,

Fix bug in percentage increase computation for cache sensitivity analysis,
add additional field trial values for cache sensitivity analysis.
R=rvargas@chromium.org
Committed: https://src.chromium.org/viewvc/chrome?view=rev&revision=160004

[rvargas (doing something else), 2012-10-03 02:01:09]

http://codereview.chromium.org/11045008/diff/1/chrome/browser/chrome_browser_...
File chrome/browser/chrome_browser_field_trials.cc (right):

http://codereview.chromium.org/11045008/diff/1/chrome/browser/chrome_browser_...
chrome/browser/chrome_browser_field_trials.cc:576: base::FieldTrial::Probability
sensitivity_analysis_probability = 8;
50% of canary is delayed? can we lower this percentage?

http://codereview.chromium.org/11045008/diff/1/net/http/http_cache_transactio...
File net/http/http_cache_transaction.cc (right):

http://codereview.chromium.org/11045008/diff/1/net/http/http_cache_transactio...
net/http/http_cache_transaction.cc:153: base::StringToInt(
Sorry, I'm confused now. What was the bug?. Why two groups for every value?

[tburkard, 2012-10-03 16:47:21]

https://codereview.chromium.org/11045008/diff/1/chrome/browser/chrome_browser...
File chrome/browser/chrome_browser_field_trials.cc (right):

https://codereview.chromium.org/11045008/diff/1/chrome/browser/chrome_browser...
chrome/browser/chrome_browser_field_trials.cc:576: base::FieldTrial::Probability
sensitivity_analysis_probability = 8;
Eventually, yes.  However, at this point, I need more data.  The data coming in
so far already shows that the impact on users is negligible (~2% slowdown), so
this is not a big deal in terms of user impact.  I want to leave this in Dev for
a week to get better data, then I'll reduce it.  SG?

https://codereview.chromium.org/11045008/diff/1/net/http/http_cache_transacti...
File net/http/http_cache_transaction.cc (right):

https://codereview.chromium.org/11045008/diff/1/net/http/http_cache_transacti...
net/http/http_cache_transaction.cc:153: base::StringToInt(
This change here is to make the two groups work for each value.  The reason why
we have two groups is to easily be able to spot the variance (if the
averages/percentile cutoffs for the two groups are close, low variance, if they
are far apart, high variance).  The bug was the "+ 100" below, I fixed that.

[rvargas (doing something else), 2012-10-03 22:19:05]

Please be aware of the tail. lgtm.

https://codereview.chromium.org/11045008/diff/1/net/http/http_cache_transacti...
File net/http/http_cache_transaction.cc (right):

https://codereview.chromium.org/11045008/diff/1/net/http/http_cache_transacti...
net/http/http_cache_transaction.cc:153: base::StringToInt(
On 2012/10/03 16:47:21, tburkard wrote:
> This change here is to make the two groups work for each value.  The reason
why
> we have two groups is to easily be able to spot the variance (if the
> averages/percentile cutoffs for the two groups are close, low variance, if
they
> are far apart, high variance).  The bug was the "+ 100" below, I fixed that.

I thought the original behavior is what you wanted at that time... Anyway, OK,
if we have different results from both groups that's a strong signal that
something is not working :)

[jam, 2012-10-03 23:14:22]

lgtm