103745 - evaluate performance of mozilla cookies implementation

Status: RESOLVED FIXED

(Core :: Networking: HTTP, defect, P3)

Description

[Darin Fisher]

evaluate performance of mozilla cookies implementation.  in particular, i'd like
to determine how expensive it is to be calling into the cookie manager for every
http request/response.

Comment 1

[Darin Fisher]

here's some results using a linux commercial build from 2001/10/01:

[with libcookie.so]
Test id: 3BC15AD47D
Avg. Median : 1060 msec     Minimum     : 252 msec
Average     : 1092 msec     Maximum     : 3812 msec
http://cowtools.mcom.com/page-loader/graph.pl?id=3BC15AD47D

[without libcookie.so]
Test id: 3BC15CFDD8
Avg. Median : 1027 msec     Minimum     : 247 msec
Average     : 1051 msec     Maximum     : 3826 msec
http://cowtools.mcom.com/page-loader/graph.pl?id=3BC15CFDD8

this amounts to a 3% performance difference, and i think it means that we
should not be sending notifications to the cookie manager unless we actually
have cookies to report.  see bug 96178 for existing plans to make this possible.

Comment 2

[Darin Fisher]

morse: this bug might interest you.

Comment 3

[Darin Fisher]

something along the lines of bug 96178 might help.

Comment 4

[Darin Fisher]

-> 0.9.8 (perhaps even later)

Comment 5

[Darin Fisher]

unlikely to get around to doing anything about this in the mozilla1.0 timeframe.

-> future

Comment 6

[Andrew Taylor]

Is this bug still relevent now that the cookie code is being rewritten?

Comment 7

[Darin Fisher]

sort of.. this is more of a meta bug than anything else.  the point of this bug
was to determine why the presence of the cookie DLL was so costly.  we may or
may not have addressed all of that with the cookie rewrite.  i doubt it since we
didn't see much of any change in Tp with the current cookie changes :-/

Comment 8

[[e-mail]]

ah, thx for cc'ing me darin.

"cookies" and "performance" do not go in the same sentence, unless there's a
strong negation present. the rewrite has not done anything to address
performance yet (apart from micro optimizations and other fun stuff).

the perf improvements will be coming up shortly. phase 3 sets the stage for
this, and I may introduce arenas there (they're so easy) so that'll give us some
malloc bloat + perf wins.

the really big win will come when
a) we move to hashes (phase 4, dare i mention 1.4b in this sentence? ;)

darin: i've actually just finished coding up the hash conversion. so do let me
know if there's a chance of it being on the radar for 1.4b, and i'll push it
through ASAP. not sure on this one because although it's a conceptually simple
change, it is a big patch and i'm not sure what your take will be for 1.4b.

b) we integrate cookies into necko and kill that infernal nsCookieHTTPNotify.

Comment 9

[benc]

dwitte: any update?

Comment 10

[Marco Castelluccio [:marco]]

Should this bug still be opened?

Comment 11

[[e-mail]]

Nope. Cookies are fast now!