Right now e10s is about 25% worse on Talos tpaint (on Aurora). https://treeherder.mozilla.org/perf.html#/graphs?series=[mozilla-aurora,a989bf8a925eecb3d2d508ec0dda78588a781b13,1]&series=[mozilla-aurora,d0a85e9de2bec8153d2040f2958d979876542012,1] However, telemetry seems to show that e10s opens windows more quickly. https://telemetry.mozilla.org/new-pipeline/dist.html#!compare=e10sEnabled&cumulative=1&end_date=2016-02-20&keys=__none__!__none__!__none__&max_channel_version=aurora%252F46&measure=FX_NEW_WINDOW_MS&min_channel_version=null&product=Firefox&sanitize=1&sort_keys=submissions&start_date=2016-01-28&table=1&trim=1&use_submission_date=0 We need to figure out the source of this disagreement. Joel, do you have any ideas here?

lets look at mozilla-central also: https://treeherder.mozilla.org/perf.html#/graphs?series=[mozilla-aurora,a989bf8a925eecb3d2d508ec0dda78588a781b13,0]&series=[mozilla-aurora,d0a85e9de2bec8153d2040f2958d979876542012,0]&series=[mozilla-central,a989bf8a925eecb3d2d508ec0dda78588a781b13,1]&series=[mozilla-central,d0a85e9de2bec8153d2040f2958d979876542012,1] that shows the same pattern. the source for tpaint: https://dxr.mozilla.org/mozilla-central/source/testing/talos/talos/startup_test/tpaint.html we launch a new window (CTRL+N, but done with prefs so we don't open a new tab) and load a script, this waits for mozafterpaint event on an empty page and sends a message back to the parent window. I am not sure what telemetry is measuring, but filtering down telemetry to win7, I see this https://telemetry.mozilla.org/new-pipeline/dist.html#!arch=x86!x86-64&compare=e10sEnabled&cumulative=1&end_date=2016-02-20&keys=__none__!__none__!__none__&max_channel_version=aurora%252F46&measure=FX_NEW_WINDOW_MS&min_channel_version=null&os=Windows_NT%252C6.1&product=Firefox&sanitize=1&sort_keys=submissions&start_date=2016-01-28&table=1&trim=1&use_submission_date=0 My experience with interpreting telemetry data is very limited, but I see e10s showing much higher numbers in all buckets, about the same ratio of 1 non-e10s to 7-10 e10s data points, including up to the high bucket (infinity). This indicates loosely that the time to open a new window is the same- is there a different way to interpret the data?

Regarding the telemetry, here's how to interpret the data. The numbers themselves aren't relevant because there are different numbers of people using e10s and non-e10s. So you have to look at the percentages. Let's read across one line, which (for me) looks like: Start End e10s non-e10s 472 785 25.17k (90.63%) 106.27k (86.12%) This line means "90.63% of e10s users were able to open a new window in <= 785ms. 86.12% of non-e10s users were able to open a new window in <= 785ms." If you look at each line, you can see that the e10s percentages are always higher (except at infinity). This means that more e10s users were able to get a new window within that time than non-e10s users (percentage-wise).

thanks for explaining that. It does look as though e10s would be faster. I guess now we need to figure out where the telemetry data gets the time and what exactly that means. looking on dxr, I see: https://dxr.mozilla.org/mozilla-central/source/browser/base/content/browser.js#3835 that appears to be that we are waiting for the message "document-shown" for marking it. That is a lot different than mozafterpaint. A few thoughts for an interested hacker: 1) collect telemetry from a talos run to see if e10s vs opt yields similar telemetry to what we see in the wild 2) test talos tpaint with 'document-shown' instead of mozafterpaint, not sure how exactly to do that, but I assume it could be hacked up.

We've been doing a lot of digging into talos tests, in particular tpaint and trust the numbers we have there. Downgrading this since we're not tracking FX_NEW_WINDOW_MS.

A quick look at FX_NEW_WINDOW_MS, and it looks like this only records a time delta when the window is opened by the parent. This is something I determined was not regressed by e10s in bug 1251732. This explains the discrepancy, as tpaint opens the new window from the content process, by design. I think we're done here.