Call for testing for 3.5 OpenSSH

[Dan Kaminsky]

Kevin Steves wrote:

>On Mon, Sep 23, 2002 at 12:46:08PM -0700, Dan Kaminsky wrote:
>  
>
>>   Horrendously laggy console -- totally unusable.  SO_NODELAY get nuked?
>>    
>>
>
>Nagle usage is unchanged.  This problem report is useless.
>  
>
It's called a problem report, not a solution report.  Given that:

1) Testing calls have preceded releases by 24 hours
2) Testing calls that go unheeded tend to lead to a very unhappy Ben
3) Ben specifically asked about Cygwin
4) I run Cygwin

I thought it was prudent to at least mention "compiles fine, not 
actually usable though" as quick as possible.

Here, lemme see if I can whip up something a bit more useful.  I don't 
know exactly what's causing it, but there's a second-level timeout 
somewhere in here.  I fired up tethereal, and monitored the inter-packet 
timings on a standard SSH2 login.  Here's the absolute timing -- notice 
time-to-password is about 3 seconds, but in CVS we're up to 12.

Normal OpenSSH Behavior on initial connect(cygwin to freebsd):
19.787118
19.787173
19.787412
19.788726
19.800843
19.802972
19.803673
19.898120
19.898436
19.911058
19.951071
20.020607
20.084571
20.178110
20.178442
20.178605
20.181870
20.182947
20.189624
20.191384
20.197101
20.197232
20.339739
22.408260
22.408285
22.408596
22.408829

Broken CVS OpenSSH on same:
 2.801344
 2.801393
 2.801614
 2.802872
 2.822554
 2.825399
 2.827688
 2.917852
 3.550563
 3.562710
 3.689510
 4.357904
 4.423234
 4.592200
 5.129863
 5.227887
 5.228141
 5.228311
 5.394639
 5.841215
 5.842401
 5.996435
 6.927709
 6.929665
 7.099771
 7.813722
 7.813926
 8.002521
14.456602
14.456629
14.456945
14.457173

Looks like socket vomit to me, Kevin.

Log attached.

--Dan

-------------- next part --------------
A non-text attachment was scrubbed...
Name: packetlog
Type: application/octet-stream
Size: 25140 bytes
Desc: not available
Url : http://lists.mindrot.org/pipermail/openssh-unix-dev/attachments/20020923/82d96f08/attachment.obj