146950 - [osx] 1.0rc3 browser is incredibly slow using SSL

Status: VERIFIED DUPLICATE of bug 147979

(Core :: Networking: HTTP, defect)

Description

[Robert Walsh]

The 1.0rc3 browser is really slow on MacOS X.  So slow, in fact, that I'm going
to revert to 1.0rc1, which worked just fine.

Specifically, it takes a really long time to connect or download data from a web
page - even ones that are on my LAN.  All other applications that use the
network run just fine, as does rc1, so I know it's nothing to do with the network.

Comment 1

[Matthias Versen [:Matti]]

-> Networking:http 

Comment 2

[Greg K.]

WorksForMe using FizzillaCFM/RC3.

Robert, can you reproduce this slowness using a new Mozilla user profile?

Comment 3

[Robert Walsh]

I'll certainly try, later on tonight when I get back home.  More details then.

Comment 4

[Robert Walsh]

No joy - I created a new profile and am seeing the exact same problem.  Works
fine from other machines on my network, and from other applications (like IE) on
this machine.

Comment 5

[Darin Fisher]

reporter: what version of OSX are you using?

Comment 6

[Robert Walsh]

I am using 10.1.4, with all of the latest updates from Apple.  The machine is an
ibook with 256MB of memory and I use an airport card for connectivity.  I've
eliminated the system software and hardware as being the problem by running IE
and Mozilla 1.0rc1.  Both worked just fine.  I even ran IE and 1.0rc3 at the
same time to spot the time difference.

Specifically, when I access my work squirrelmail account, it takes IE about 3
seconds to go from the log in screen to the complete set of main screen frames.
 It takes Mozilla 1.0rc3 about 3 seconds to display the first frame and about 20
seconds to display the second frame.  1.0rc1 is about as fast as IE.

Comment 7

[Steve Dagley]

Any chance the problematic websites or the 2nd frame of the squirrelmail site
are  secure pages?  I've gotten some e-mail from folks complaining that some
sites with their own CAs have started loading v e r y  s l o w l y.

Comment 8

[Robert Walsh]

Well, both frames are secure.  Actually, I've noticed this happening with the
right-side frame, too, but usually it's such a small amount of data it doesn't
happen.

As for the CA issue - I've tried this on two SquirrelMail web sites, one with
its own CA (my own one at home) and one with a VeriSign CA (my mail account at
work.)  Both exhibit the same slow behaviour, so I don't think the CA being a
well-known CA has anything to do with it.

Comment 9

[Darin Fisher]

kaie, ssaux: here's another bug about secure sites loading very very slowly.

Comment 10

[Greg K.]

See also bug 147979.

Comment 11

[Robert Walsh]

FYI: this is still happening with the 1.0 final release.  It appears to be
effecting https only.  I've also upgraded to Mac OS X 10.1.5 and the problem is
still happening.

Comment 12

[Mike Pinkerton (not reading bugmail)]

this is a blocker for embed clients.

Comment 13

[Hong Kwon]

We will try to setup an SSL version of pageloader to get some data.

Comment 14

[Stephane Saux]

cc javi
Javi thinks it may be a regression in the dbm code on the mac platform.

Comment 15

[John Morrison]

The pageload test on 'cowtools' will now run with either https or http (you 
can select which flavour to run by clicking on the appropriate URL). It does
use a 'self-certified' certificate. (Do we have a way to get a certificate
with a proper CA signing?). https://cowtools.mcom.com/page-loader/loader.pl

I haven't tried this on OS/X, but the current trunk on win2k doesn't show 
significant speed loss in the SSL case. [However, we may want to get this 
setup so that we can test over modem, since poor SSL performance would be
accentuated for a low-bandwidth/high-latency connection, no?].

Comment 16

[Robert Walsh]

This doesn't surprise me - it also works fine on 1.0 final on Linux, so it
definitely appears to be OS X specific.

Comment 17

[John Morrison]

Man. Mea culpa. I should have been all over this.

This isn't just "slow". It's completely busted. This is up to 10 times slower
in current builds than in RC1 when the test doesn't time out.

I went back through the daily 2002-mm-dd-hh-1.0.0/mozilla-macosX-1.0.0.smi.bin 
builds on sweetlou. SSL runs okay in 2002-04-27-05-1.0.0, 2002-04-28-05-1.0.0,
and 2002-04-29-05-1.0.0 builds, but chokes on 2002-04-30-05-1.0.0 and 
2002-05-01-05-1.0.0 builds. So something bad was checked in during the 4/29am
to 4/30am window (see url on this bug). 

I see some NSS related code in that window. There's also darin's pipelining
changes, although those should be pref-ed off. And there are a rew mac specific
changes, but they look innocuous.

(tested on os/x 10.1.5 733MHz 256MB G4)

Comment 18

[Wan-Teh Chang]

Attachment: NSS, NSPR, and mozilla/dbm changes between 1.0rc1 and 1.0rc3

The NSS changes shown in the URL on this bug were not
made to the MOZILLA_1_0_0_BRANCH used by 1.0rc1 and
1.0rc3.

This patch is the Mac related changes in NSS and its
dependencies (NSPR and mozilla/dbm) between 1.0rc1 and
1.0rc3.

Comments:
1. There are no NSPR changes.
2. The NSS change simply exports one more function from
SMIME3.shlb.
3. The DBM change is substantial and is the fix for bug
129080.  It was committed at 27 Apr 2002 03:43:49 -0000,
not during the 4/29am to 4/30am window noted in comment
#17.

Comment 19

[John G. Myers]

*** This bug has been marked as a duplicate of 147979 ***

Comment 20

[Tom Everingham]

verified duplicate