151762 - perf on escaping and unescaping a string

Status: RESOLVED INACTIVE

(Core :: DOM: Core & HTML, defect, P5)

Description

[Markus Hübner]

Using a 1.1ghz,512ram workstation I get the following restults

trunk build 2002061308: 1092 ms
MSIE 6: 160 ms
NS 4.78: 521 ms

This performance issues lead to some problems as stated in n.p.m.performance 
posting: 
"The escaping and unescaping a string (URL) is another daunting one. In order 
to be able to have Moz's dom parser parse our SOAP's return we have to escape 
on server, unescape on client for each call... no wonder the xmlhttp object 
seems so slow on Moz, the stinky delay of unescaping."

Comment 1

[Markus Hübner]

btw - the newsgroup posting is here:
http://groups.google.com/groups?
hl=de&lr=&ie=UTF8&oe=UTF8&th=3c3a1e0d0ec9727&rnum=1

Comment 2

[Johnny Stenback (:jst)]

Do we have testcases and numbers that show that the JS functions escape/unescape
are to blame for a large part of the performance problem here?

Comment 3

[Thad Hoffman]

I don't have any accessible real life test cases due to firewalls and appserver
licenses, but  I can offer up a series of SOAP payloads that our system uses to
time escaping/unescaping them.

Comment 4

[Thad Hoffman]

I have a page that escapes the unescapes the same strings.
 Running tests on a win 2000 advanced server dell latiture c800 pIII 866MHz
256Megs Ram and a Mac G4 400Mhz OS 10.1.5 448Megs of Ram.

(note: repeated tests 25 times to see differences)

IE 5.1.4 (mac os X) 
escape: 6ms
unescape: 9ms

IE 5.5 sp2 (win 2k)
escape: <1ms (every 4th time was 10ms)
unescape: <1ms

Moz 1.0 (Mac OS X)
escape: 14ms(fastest) 40ms(slowest) (same numbers pop up, 14, 18, 34, 40 not
consistent results)
unescape: 6-7ms 

Moz 1.0 (Win 2k)
escape:<1ms (every 3rd 10ms)
unescape: <1ms (every 3rd 10ms)

These results are with our smallest SOAP returns. I'm gathering our larger ones
to test against. So far my findings aren't exactly in line with the perf results
posted at: http://bugzilla.mozilla.org/attachment.cgi?id=87424&action=view

Comment 5

[Markus Hübner]

reassigning ...

Comment 6

[Markus Hübner]

Any chance we get the patch in for 1.2 ?

Comment 7

[John Keiser (jkeiser)]

I am working on some string stuff that could help perf in string conversion
routines, but I am pretty overloaded for 1.2.  If someone else wants to take
this, feel free.  If not I'll take a look when I've done with the other string
stuff.

Comment 8

[John Keiser (jkeiser)]

I can't say when I'll get back to CharFilter.  Other major things are in my way
for a long while.  Others should feel free to work on this.

Comment 9

[Johnny Stenback (:jst)]

Mass-reassigning bugs to [e-mail]

Comment 10

[Markus Hübner]

jkeiser, are you willing to take this?

Comment 11

[John Keiser (jkeiser)]

Not for a while, no.

Comment 12

[Wayne Mery (:wsmwk)]

still a problem?