read/write multiplexed support

[Marcel Müller]

Christopher R. Hertel wrote:
> It would also help to have the NegProt.  Since the dialect is prior to NT LM
> 0.12 we won't see capabilities bits, but we will see whether the Raw Mode
> bits are set in the NegProt response.

I think you are right.

There are two scenarios.
1. The connection is slow from the beginning.
2. The connection is initially fast and slows down after a few hours.

In the second case there was always a timeout involved in some way. So I 
think we can forget about that, since the timeouts are only reproducible 
with a certain version of the IBM requester, which is not the most 
recent one.

But I conclude from that, that the speed (and the transfer protocol) 
does not change once the session is active.

But the first case is curious. I only have a vague guess. The disk in 
the server might be down at when the client connects in the first time. 
The spin up may take too long for some SMB request, again resulting in a 
timeout. Unfortunately I was not able to reproduce this by stopping the 
device explicitly with sdparm. But the data required at the session 
setup may still be in the cache, when the disk is down. Furthermore it 
is not that easy to do a tcp dump while the disk is down.

I now did some tweaks to the client configuration. In fact I disabled 
the multiplex modes in the wrkheuristics setting. This should at least 
force the differences between samba 3.0.x and samba 3.2.x to disappear.
Since that, I was not able to reproduce the problem with samba 3.2.5 
with about 5 longer sessions. No Idea why. Maybe this happened only by 
chance. Maybe the client takes the raw and the multiplexed modes coequal 
and to disable one of them fixed it. This would also explain, why the 
communication with windows servers is always fast, because they never 
support both modes at once.

I will let you know if I have something new.
As far as I see the solution will be in the surrounding of the 
negotiation, if any.

Btw, is it possible that the server still claims to support multiplexed 
modes at negotiation?


Marcel