graph server shows a regression: http://graphs.mozilla.org/graph.html#tests=[[255,131,31]]&sel=1421252757780.488,1421392343530.2808,211.47540983606558,275.40983606557376&displayrange=30&datatype=geo I did some retriggers on tbpl: https://treeherder.mozilla.org/#/jobs?repo=mozilla-inbound&fromchange=a3d8163ee939&tochange=a091ec8d2a04&filter-searchStr=Windows 8 64-bit mozilla-inbound talos tp5o and this changeset is where we went from tp5 having values of ~234.0 -> ~240.0: https://hg.mozilla.org/integration/mozilla-inbound/rev/f925d9830745

:hurley, can you comment on if this is expected or unexpected? I would like to see us come to a decision no later than Thursday on a course of action to either: * backout * fix this * accept the regression based on the value from the fix * have a plan with a timeline for addressing this for reference here is the try syntax to run on try: try: -b o -p win64 -u none -t tp5o

(In reply to Joel Maher (:jmaher) from comment #1) > :hurley, can you comment on if this is expected or unexpected? I would like > to see us come to a decision no later than Thursday on a course of action to > either: > * backout > * fix this > * accept the regression based on the value from the fix > * have a plan with a timeline for addressing this I'd add also: accept without further investigation because <A, B, C>

Joel, this is expected, based on previous experiences with this feature (though with a different backend). You can see the long explanation of everything in my original post here https://groups.google.com/forum/#!msg/mozilla.dev.platform/tAnuc-J0DbY/snrGsOVW1MAJ Additionally, I've been doing profiling of this feature locally to ensure we aren't running up against any of the issues we've had with previous iterations of this feature, and my profiling is all coming up great so far. I still want to run a few more tests with some other configurations once I get back in my home office, but everything looks significantly better this time around than it did last time. As such, I'm particularly worried about this 3% regression. I am curious, though, is windows 8 the *only* platform that regressed on tp5? Last time, the regressions were on all platforms, ranging from 6% to 8%.

ok, great question. I looked in more detail and this is on win8, win7, and winxp. Windows 8 is the most pronounced: http://graphs.mozilla.org/graph.html#tests=[[255,131,31],[255,131,37],[255,131,25]]&sel=1421205200511.6072,1421471487172.7507,147.5409836065574,354.0983606557377&displayrange=30&datatype=geo There is a tp5 regression a day earlier which affects linux* and osx10.8 with mild influence on win*. So this is expected, now that this is confirmed on win*, is that matching your expectations. If so, shall we close this as wontfix ?

I would like to confirm that this is all expected. If so I would be happy to close this as wontfix

Joel, thanks. Yes, I think we can close this as wontfix. Regardless of this bug, I'm keeping a close eye on anything related to the predictor landing anyway, given my previous experience with it and a chemsipll :)