New issue
Advanced search Search tips
Note: Color blocks (like or ) mean that a user may not be available. Tooltip shows the reason.

Issue 649595 link

Starred by 5 users

Issue metadata

Status: WontFix
Owner: ----
Closed: Sep 2016
Cc:
Components:
EstimatedDays: ----
NextAction: ----
OS: Windows
Pri: 2
Type: Bug-Regression



Sign in to add a comment

Canary v55 seems to be slower than stable v53

Reported by insensi...@gmail.com, Sep 23 2016

Issue description

UserAgent: Mozilla/5.0 (Windows NT 10.0; Win64; x64) AppleWebKit/537.36 (KHTML, like Gecko) Chrome/55.0.2868.0 Safari/537.36

Steps to reproduce the problem:
1. Visit http://infernojs.org/benchmarks/angular-test-table/infernojs/index.html
2. Select 'InfernoJS' and click '5000 records'

What is the expected behavior?

What went wrong?
On my machine, average time shown on this page is ~190ms on Chrome 53 and ~220ms on Chrome 55 Canary. First I seen this performance degradation on my own closed-source project with custom HTML rendering engine. 

Did this work before? Yes In Chrome Version 53.0.2785.116 m

Chrome version: 55.0.2868.0  Channel: canary
OS Version: 10.0
Flash Version: Shockwave Flash 23.0 r0

To reproduce this issue page and get clean results the page http://infernojs.org/benchmarks/angular-test-table/infernojs/index.html should be reloaded every time before '5000 records' button is clicked.
 

Comment 1 by tkent@chromium.org, Sep 23 2016

Labels: Needs-Bisect
Cc: tkonch...@chromium.org
Labels: Needs-Feedback
Tested the same on win10 chrome version canary 55.0.2871.0 and stable 53.0.2785.116 - Observed that stable takes 125 ms whereas canary takes 131 ms

Could you please recheck the same on a new profile and update the thread with your observations.
649595.png
99 KB View Download
Components: Blink>DOM

Comment 4 by woxxom@gmail.com, Sep 26 2016

Win7x64.

1st number: reloading the page + "5000 records"
2nd number: "5000 records", wait, "5000 records" without reloading the page

45.0.2454.101 stable: 142 ms, ~90ms
53.0.2785.101 stable: 150 ms, ~75ms
55.0.2868.3 dev:      150 ms, ~70 ms
55.0.2872.0 canary:   143-181 ms, ~75ms

So Canary's first run time varies more than in previous builds, sometimes faster, sometimes slower than stable.
In my opinion, the most important part here, is 1st number - "reloading the page + "5000 records". As shown, dev/stable have highest bound 150ms, but Canary has 181 ms which is ~10-15% slower than dev/stable.
Today I did more research on this issue. Here are my observations:
1. This problem may be not related to DOM operations at all. On my TypeScript project with custom Virtual-DOM like rendering engine I see performance degradation even on render stages, which not use DOM at all.
2. I ran Octane JavaScript benchmark and see, that some tests are significantly slower in Canary, mostly "Regexp" and "TypeScript" tests.
I used for tests:

Stable 53.0.2785.116, 
Canary 55.0.2873.0 
Canary 55.0.2873.3 (received an update today).

All screenshots for Octane tests in the attachement. Note, that "TypeScript" tests seems to run faster in Canary 55.0.2873.3.
Octane_Stable_53.0.2785.116.png
85.9 KB View Download
Octane_Canary_55.0.2873.0.png
87.8 KB View Download
Octane_Canary_55.0.2873.3.png
85.1 KB View Download

Comment 7 by tkent@chromium.org, Sep 28 2016

Components: -Blink>DOM Blink>JavaScript
Labels: -Needs-Feedback
Summary: Canary v55 seems to be slower than stable v53 (was: Canary v55 seems to be slower with DOM operations than stable v53)
I just made performance test on my own application/framework in latest stable and latest canary. It seems that updated today Canary 55.0.2873.3 runs faster than previous release and performance is closer to stable. I am unable to provide source code for these tests, there are only screenshots if they can be useful.
my_benchmark_stable_53.0.2785.116.png
23.3 KB View Download
my_benchmark_canary_55.0.2873.3.png
17.4 KB View Download
V8 is currently a/b testing an interpreter on Dev/Canary. It is very likely that you are in the interpreter group which is a little bit slower on peak performance benchmarks. In order to verify this please do the following thing:

1.) Open Canary
2.) Goto about:flags
3.) Set "Experimental JavaScript Interpreter" to "Disabled"
4.) Rerun your benchmark.
Thank you for a suggestion. I disabled 'Ignition' interpreter my observations are:
1. When I tried to run my application in Canary, I got 'stack overflow' javascript error. Canary version is now '55.0.2874.0', yesterday I ran my tests in Canary '55.0.2873.3'. I experimented with 'Ignition' state (Disabled, Enabled, Default) - same stack overflow error in my application. Works fine in stable, and worked fine yesterday in Canary '55.0.2873.3'
2. When I ran Octane benchmarks I saw no difference when I set 'Ignition' to 'Default' or 'Disabled'. I got different results only when I set 'Ignition' to 'Enabled' so I assume I not have it enabled by default..

Hope this helps. Maybe it makes sense to run Octane benchmarks and experiment with 'Ignition' (maybe other Canary flags?) from your side?
Labels: Needs-Feedback
Tested the same on win10 chrome version 55.0.2875.0 and stable 53.0.2785.116 with reloading the page + "5000 records" observed 144ms on canary and 151ms on stable version.

Following the steps in comment #9 with reloading the page + "5000 records" observed 136ms on canary and 132ms on stable version
 
insensible@ Could you please look into the result obtained and update your observations
I performed original test at  http://infernojs.org/benchmarks/angular-test-table/infernojs/index.html and saw almost exactly the same execution timings ~200ms in Stable 53.0.2785.143 and Сanary 55.0.2875.0.

But in benchmarks in my application still see performance degradation between Stable/Canary, and some performance degradation can be seen on Octane benchmarks.

So original issue is now not reproducible, maybe this task should be closed..

Hope Canary JS engine will be faster even on most projects after final release.
Status: WontFix (was: Unconfirmed)
WontFix according to #11 and #12

Sign in to add a comment