110408 2013-02-20 16:49:44 -0800 Tune BackgroundHTMLParser's pendingTokenLimit based on a benchmark 2013-02-21 13:48:14 -0800 1 1 1 Unclassified WebKit WebCore Misc. 528+ (Nightly build) Unspecified Unspecified RESOLVED FIXED P2 Normal --- 106127 1 tonyg tonyg abarth eric esprehn+autocc ojan.autocc webkit.review.bot oldest_to_newest 837933 0 tonyg 2013-02-20 16:49:44 -0800 I ran the threaded HTML parser in Telemetry against the top 25 sites on a Nexus 7 with a variety of pendingTokenLimits ranging from 250-6000 (current == 4000). My observations: - The threaded HTML parser improves by DOMContentLoaded 8-14% and total time spent parsing on main thread by 9-13% - Varying pendingTokenLimit doesn't significantly change total time spent parsing on main thread, DOMContentLoaded or Load - Tweaking pendingTokenLimit can improve maximum time spent parsing on main thread by another 40% I recommend we change it from 4000 to 1000. Results: https://docs.google.com/a/chromium.org/spreadsheet/ccc?key=0AmVDuVhIZxCTdEstXy02NXZnNE5wb0d4NTlPNGQ2U1E#gid=1 837958 1 eric 2013-02-20 17:07:09 -0800 Discussed with Tony in person. I'm worried that but 109995 could be throwing off our perf numbers, and we should probably wait to do final testing for this change until after bug 109995 is resolved. It would also be nice to see numbers for the "yield every 10 tokens" case, just to verify that it throws out our "total parse time" (or one of our metrics) by an expectedly large number. I worry our current inspector-based testing doesn't account for the overhead this contant is trying to avoid. :) 838588 2 tonyg 2013-02-21 11:00:46 -0800 I ran with pendingTokenLimit=10 and updated the spreadsheet. Both total parse time and DOMContentLoaded markedly regressed. 838597 3 189560 tonyg 2013-02-21 11:08:15 -0800 Created attachment 189560 Patch 838670 4 189560 abarth 2013-02-21 12:10:46 -0800 Comment on attachment 189560 Patch This patch seems harmless. We might want to tune again when we're closer to turning the threaded parser on by default. 838793 5 189560 webkit.review.bot 2013-02-21 13:48:10 -0800 Comment on attachment 189560 Patch Clearing flags on attachment: 189560 Committed r143649: <http://trac.webkit.org/changeset/143649> 838794 6 webkit.review.bot 2013-02-21 13:48:14 -0800 All reviewed patches have been landed. Closing bug. 189560 2013-02-21 11:08:15 -0800 2013-02-21 13:48:10 -0800 Patch bug-110408-20130221110435.patch text/plain 1910 tonyg U3VidmVyc2lvbiBSZXZpc2lvbjogMTQzNTM4CmRpZmYgLS1naXQgYS9Tb3VyY2UvV2ViQ29yZS9D aGFuZ2VMb2cgYi9Tb3VyY2UvV2ViQ29yZS9DaGFuZ2VMb2cKaW5kZXggZTgwZDYzZjAyMGQ3OGIz ZTA1NTg0MzZhZDhiZjRiOGUyNjNiOTk3Yy4uNzY0ODNkMzRhYmQ1MGU1MTNkMzQ1ZjM4MDJkZGJj NDQ3OGE2ODBhNCAxMDA2NDQKLS0tIGEvU291cmNlL1dlYkNvcmUvQ2hhbmdlTG9nCisrKyBiL1Nv dXJjZS9XZWJDb3JlL0NoYW5nZUxvZwpAQCAtMSwzICsxLDE4IEBACisyMDEzLTAyLTIxICBUb255 IEdlbnRpbGNvcmUgIDx0b255Z0BjaHJvbWl1bS5vcmc+CisKKyAgICAgICAgVHVuZSBCYWNrZ3Jv dW5kSFRNTFBhcnNlcidzIHBlbmRpbmdUb2tlbkxpbWl0IGJhc2VkIG9uIGEgYmVuY2htYXJrCisg ICAgICAgIGh0dHBzOi8vYnVncy53ZWJraXQub3JnL3Nob3dfYnVnLmNnaT9pZD0xMTA0MDgKKwor ICAgICAgICBSZXZpZXdlZCBieSBOT0JPRFkgKE9PUFMhKS4KKworICAgICAgICBUaGlzIGNvbnN0 YW50IHdhcyB0dW5lZCBieSBydW5uaW5nIHRoZSB0b3AgMjUgc2l0ZXMgaW4gVGVsZW1ldHJ5IG9u IGEgTmV4dXMgNyBkZXZpY2UuCisgICAgICAgIFRoZSBuZXcgdmFsdWUgZGVjcmVhc2VzIHRoZSBt YXhpbXVtIHRpbWUgc3BlbnQgcGFyc2luZyBieSA0MCUgd2l0aG91dCBzaWduaWZpY2FudGx5IGlt cGFjdGluZyB0b3RhbCBwYXJzZSB0aW1lIG9yIERPTUNvbnRlbnRMb2FkZWQuCisKKyAgICAgICAg Tm8gbmV3IHRlc3RzIGJlY2F1c2Ugbm8gbmV3IGZ1bmN0aW9uYWxpdHkuCisKKyAgICAgICAgKiBo dG1sL3BhcnNlci9CYWNrZ3JvdW5kSFRNTFBhcnNlci5jcHA6CisgICAgICAgIChXZWJDb3JlKToK KwogMjAxMy0wMi0yMCAgQWxlYyBGbGV0dCAgPGFsZWNmbGV0dEBjaHJvbWl1bS5vcmc+CiAKICAg ICAgICAgTGV2ZWxEQjogUmVtb3ZlIGV4Y2VzcyB2ZWN0b3IgY29weQpkaWZmIC0tZ2l0IGEvU291 cmNlL1dlYkNvcmUvaHRtbC9wYXJzZXIvQmFja2dyb3VuZEhUTUxQYXJzZXIuY3BwIGIvU291cmNl L1dlYkNvcmUvaHRtbC9wYXJzZXIvQmFja2dyb3VuZEhUTUxQYXJzZXIuY3BwCmluZGV4IDdkODM4 ZDEyNGQ3YTYyZmQxOTQ3MWJhNzk5ZDZmNTJmOTIwZTc2MTYuLjU3MTdlYzEzY2Y1NGU0YjUyZjEw NDMxYTJlNjhlZDgyMTQzNTg0OWYgMTAwNjQ0Ci0tLSBhL1NvdXJjZS9XZWJDb3JlL2h0bWwvcGFy c2VyL0JhY2tncm91bmRIVE1MUGFyc2VyLmNwcAorKysgYi9Tb3VyY2UvV2ViQ29yZS9odG1sL3Bh cnNlci9CYWNrZ3JvdW5kSFRNTFBhcnNlci5jcHAKQEAgLTExMiw4ICsxMTIsNyBAQCBzdGF0aWMg aW5saW5lIGJvb2wgdG9rZW5FeGl0c0ZvcmVpZ25Db250ZW50KGNvbnN0IENvbXBhY3RIVE1MVG9r ZW4mIHRva2VuKQogICAgICAgICB8fCAodGhyZWFkU2FmZU1hdGNoKHRhZ05hbWUsIGZvbnRUYWcp ICYmICh0b2tlbi5nZXRBdHRyaWJ1dGVJdGVtKGNvbG9yQXR0cikgfHwgdG9rZW4uZ2V0QXR0cmli dXRlSXRlbShmYWNlQXR0cikgfHwgdG9rZW4uZ2V0QXR0cmlidXRlSXRlbShzaXplQXR0cikpKTsK IH0KIAotLy8gRklYTUU6IFR1bmUgdGhpcyBjb25zdGFudCBiYXNlZCBvbiBhIGJlbmNobWFyay4g VGhlIGN1cnJlbnQgdmFsdWUgd2FzIGNob3NlbiBhcmJpdHJhcmlseS4KLXN0YXRpYyBjb25zdCBz aXplX3QgcGVuZGluZ1Rva2VuTGltaXQgPSA0MDAwOworc3RhdGljIGNvbnN0IHNpemVfdCBwZW5k aW5nVG9rZW5MaW1pdCA9IDEwMDA7CiAKIEJhY2tncm91bmRIVE1MUGFyc2VyOjpCYWNrZ3JvdW5k SFRNTFBhcnNlcihQYXNzUmVmUHRyPFdlYWtSZWZlcmVuY2U8QmFja2dyb3VuZEhUTUxQYXJzZXI+ ID4gcmVmZXJlbmNlLCBQYXNzT3duUHRyPENvbmZpZ3VyYXRpb24+IGNvbmZpZykKICAgICA6IG1f d2Vha0ZhY3RvcnkocmVmZXJlbmNlLCB0aGlzKQo=