168714 - Bad JS performance Moz 1.1 and 1.2 compared to Moz 1.0

Status: VERIFIED WONTFIX

(Core :: JavaScript Engine, defect)

Description

[crisp]

User-Agent:       Mozilla/4.0 (compatible; MSIE 6.0; Windows NT 5.0; Q312461)
Build Identifier: NaN

Compared to the JS performance in Moz 1.0 (which was 3 times better compared to 
IE6) the performance in Moz 1.1 and Moz 1.2 has been reduced again to IE6 
comparable levels.

Reproducible: Always

Steps to Reproduce:
1.Use Mozilla 1.0 to open page http://www.crisp.demon.nl/crackmd5.html
2.Fill in 'f3abb86bd34cf4d52698f14c0da1dc60' in MD5 and push 'Crack!'
3.On an AMD XP2000+ this takes around 80 seconds to complete
4.Use Mozilla 1.1 or 1.2 to open the same page
5.Again fill in 'f3abb86bd34cf4d52698f14c0da1dc60' in MD5 and push 'Crack!'
6.On an AMD XP2000+ it now takes around 250 seconds to complete

Actual Results:  
Mozilla 1.1 and 1.2 actually perform very poorly compared to 1.0

Expected Results:  
1.1 and 1.2 should perform equally to 1.0, or even better

Comment 1

[Dan]

With Mozilla/5.0 (Windows; U; Windows NT 5.1; en-US; rv:1.2a) Gecko/20020914 on
a Celeron/900@1124 I get 5:19.87 and with IE6 I get 4:45.12, a difference of
~11%. I don't know how fast Mozilla 1.0 is compared to IE6 though.

Comment 2

[Jo Hermans]

On build 2002091309 on Mac OS 9 (since it's a 300 MHz iBook, I don't even bother
to check how long it takes), I can see 'current password' & 'current hash'
flashing, so you script is trying to updates those on-screen fields. Could this
be the reason that it's slower ?

I thought that getElementByID() was optimized for Mozilla1.1. That might explain
why there's something changed. If you take out the screen-updates (or only do
them every 100 iterations or so), it will be a lot faster.

Comment 3

[Jo Hermans]

Oh, I got it. It's the setTimeout() call : you're using a timeout of 1 msec, and
bug 123273 has increased the minimum timeout to 10msec.

The reason why Mozilla was faster than IE, was that IE was using that minimum,
to prevent 'run away' JavaScript. Lots of pages were using a 0 msec timeout,
because they thought it means 'run as fast as possible'. Possibly true, but
sicne IE was really using 10 msec, the CPU wasn't runnign at 100%. But since
Mozilla really tried to have that 0 (or 1) msec timeout, it was very noticable
for the rest of the interface. Even though your code was really running faster
than IE ...

Sorry, I guess this a WONTFIX bug. Your code could easily be changed to execute
more 'md5cracks' in a single timeout. But we need that minimum timeout for other
pages.

Comment 4

[crisp]

Thanks y'all; I now do understand fully. Also why IE seemed to be so slow 
compared to Moz 1.0
I have adjusted my program to set the timeout every 100 tries, and came up with 
the following figures for comparison (all tests run on my AMD XP2000):

Moz 1.0: 475 pwds/sec
Moz 1.2a: 375 pwds/sec
IE6: 826 pwds/sec

Seems that Moz1.2 IS slower than 1.0, but then again it is still alpha-state. 
Would be nice to know what is causing this though.
What is more interesting is the fact that IE6 is more than twice as fast in JS 
calculation compared to Moz.
I still like Moz better though, but thought it would be interesting to let you 
know

Comment 5

[Jo Hermans]

Those figures would be more or less normal : it's true that the
JavaScript-engine of IE is a lot faster than Mozillas. Before you were seeing
the difference between the 1msec timeout (Moz) and the 10msec (IE).

I'm not sure why Mozilla 1.2a is slower that 1.0 : getElementById should be
faster. Have you tried to remove those calls out of the loop ?

Comment 6

[crisp]

getElementById is not the issue here; I am now updating the screen after every 
200 calculations, so even if that is faster in Moz 1.2 it wouldn't give any 
boost in this case. It's the javascript calculations itself that are slower in 
Moz 1.2 compared to Moz 1.0.

Comment 7

[Phil Schwartau]

crisp: thank you for this report and for your interesting site.
We will reduce the HTML from your test to make a "standalone"
JS shell testcase that can be used to test the JS Engine itelf.

cc'ing Mazie on that; we will report back with our findings -