Bug 154593
| Summary: | Huge memory usage (~10GB RAM) when running Dromaeo JavaScript Library Tests | ||
|---|---|---|---|
| Product: | WebKit | Reporter: | Carlos Alberto Lopez Perez <clopez> |
| Component: | JavaScriptCore | Assignee: | Nobody <webkit-unassigned> |
| Status: | RESOLVED WONTFIX | ||
| Severity: | Normal | CC: | bugs-noreply, cgarcia, fpizlo, kling, koivisto, mcatanzaro, ossy, rniwa |
| Priority: | P2 | ||
| Version: | WebKit Nightly Build | ||
| Hardware: | Unspecified | ||
| OS: | Unspecified | ||
Carlos Alberto Lopez Perez
I have noticed that when running the Dromaeo JavaScript Library Tests <http://dromaeo.com/?jslib> WebKit consumes a lot of memory (around [8-10]GB of RAM).
It seems that the huge increase of RAM usage happens when running the subtests "DOM Modification (jquery)".
I am able to reproduce this both with the WebKitGTK+ MiniBrowser as also with Safari on MacOS when running with WebKit from trunk (r196976).
With the WebKitGTK+ MiniBrowser from the stable branch (2.10.7 branched in r188381 <http://trac.webkit.org/wiki/WebKitGTK/2.10.x>) the memory usage is a bit better, peaks around ~[4-5]GB.
With Safari from the system (MacOS 10.10.5) the memory usage is around ~3GB.
Chromium consumes around ~3GB, and Firefox consumes "only" ~1GB.
| Attachments | ||
|---|---|---|
| Add attachment proposed patch, testcase, etc. |
Carlos Alberto Lopez Perez
This tests are ran as part of the dromaeo-jslib plan with the run-benchmark script. I noticed this huge memory usage when working on bug 153993.
So, maybe you can track this regression down with the data available on the Apple internal performance dashboard. No idea if you are also measuring the memory usage or not when running the tests.
Ryosuke Niwa
Dromaeo DOM modify is a fundamentally fraud test. There is a test that tries to add as many DOM nodes as possible (without ever removing old nodes). So the faster your JS engine, the more memory it ends up using.
Carlos Alberto Lopez Perez
(In reply to comment #2)
> Dromaeo DOM modify is a fundamentally fraud test. There is a test that tries
> to add as many DOM nodes as possible (without ever removing old nodes). So
> the faster your JS engine, the more memory it ends up using.
firefox : DOM Modification (jQuery):Runs: 569/s stdev=0.0%
chrome : DOM Modification (jQuery):Runs: 1.89K/s stdev=0.0%
safari : DOM Modification (jQuery):Runs: 4.26K/s stdev=0.0%
So yes, it seems the speed on this test is proportional to the RAM used.
Michael Catanzaro
So the higher the memory usage, the more awesome the JavaScript engine, and therefore this is not a bug? Can we close it?
Ryosuke Niwa
(In reply to comment #4)
> So the higher the memory usage, the more awesome the JavaScript engine, and
> therefore this is not a bug? Can we close it?
Yeah, I think we should close this as WONFIX.
We should really fix the test though :( It's really silly that the benchmark work like that.