On 06/13/2017 08:36 PM, Chris Peterson wrote:
Nicolas, when JSBC is enabled by default, should we change our test
procedure for our various page load tests (Talos and Softvision's manual
testing)? Since the first page load will be slower than subsequent page
loads (as you noted in the bug [1]), should we throw away the first page
load time or continue to average it with the subsequent page load times?
[1] https://bugzilla.mozilla.org/show_bug.cgi?id=900784#c72
These results [1] were with the eager encoding of the bytecode, while
running locally.
Since, I added an intuition-based heuristics which by luck end-up keeping
the encoding time as part of the ignored set, as we encode as part of the
5th visit. Thus giving the following results on tp5 [2].
Depending on how the heuristics are tuned based on telemetry results [3], we
should either split tp5 results in 2/3 sets, or increase the ignore set size.
If this is not already the case, we should change tp5 benchmark harness to
wait for the idle-callback before moving to the next page. Otherwise, we
might repeatedly attempt to encode the bytecode of one page, but never stay
long enough on the page to save it.
[1] https://bugzilla.mozilla.org/show_bug.cgi?id=900784#c72
[2] https://bugzilla.mozilla.org/show_bug.cgi?id=900784#c113
[3] https://bugzilla.mozilla.org/show_bug.cgi?id=1362114
--
Nicolas B. Pierron
_______________________________________________
dev-platform mailing list
dev-platform@lists.mozilla.org
https://lists.mozilla.org/listinfo/dev-platform
