A 60-day view of Local Twitter Page / Time to Throbber Start on http://mrcote.info/phonedash/#/rawfennecstartup shows a gradual increase. The increase looks nearly linear, with greater slope for the Nexus S phones than for the GS2. (There is no such increase for the Time to Throbber Stop.) Such a constant increment over a long period seems odd -- is there a fault in the test framework? Is something accumulating in the profiles??
Mark re-ran all the nightly builds through the s1/s2 system to verify the results. The nightlies are still showing the gradual increase. You can see the re-run of the data here: http://mrcote.info:8118/#/org.mozilla.fennec/throbberstart/local-twitter/04_46_65_fd_2f_e1_samsung_gs2|78_d6_f0_cf_8d_11_nexus_s|78_d6_f0_cf_d2_17_nexus_s/2012-05-24/2012-07-23 Here is the old data (running every pulse build - wait for it to load): http://mrcote.info/phonedash/#/org.mozilla.fennec/throbberstart/local-twitter/00_23_76_96_cc_6f_nexus_one|04_46_65_fd_2f_e1_samsung_gs2|78_d6_f0_cf_8d_11_nexus_s|78_d6_f0_cf_d2_17_nexus_s|98_0c_82_33_ec_8d_samsung_gs2|98_4b_4a_2c_e4_89_droid_pro/2012-05-24/2012-07-23
I'm not positive about this. At the moment, the framework does not reboot the phone before or after each test. I think that, perhaps, the Nexus S phones have some bug that slowly decreases performance over time. The Galaxy S2 doesn't seem affected by this. I will add a reboot before each test to see if that smooths out the line.
I am running yet another batch over the same time period (June 3 to now) but rebooting the phone before the test run. Results will be available at http://mrcote.info:8119/.
in addition to rebooting, are you running with a clean profile each time?
Yup, new profile every time. So now with reboots, there still appears to be a gradual increase in throbberstart time. Although this time it looks more like a few increases until June 17, then a levelling out. Note that throbberstop does not show this increase.
So we're pretty sure we know about the throbberstart regression in mid June, and it looks like there is no real regression after that.
Status: NEW → RESOLVED
Last Resolved: 6 years ago
Resolution: --- → INVALID
Summary: Gradual S1/S2 regression in "throbber start" time → Autophone - Gradual S1/S2 regression in "throbber start" time
You need to log in before you can comment on or make changes to this bug.