TCP stack misbehaviour?

TCP stack misbehaviour?

[clock]

The TCP stack in the linux kernel behaves this way:

I have got a FULL-DUPLEX 28.800kbps channel with BER <= 1*10-9
a) I start a long TCP connection in one direction
b) After 5 minutes I start another connection in the opposite direction

The second connection rusn for about half a minute, then converges to 0
throughtput. After several minutes, another 30 seconds of transmission ocuur.
The data path in the direction that should be used for this connection is
empty, except for occasional ACKs. The utilization of the channel is about 4%.

I would expect that both channels would be used for at least 95%. Instead, only
one is used.

Is this a bug of Linux kernel TCP stack, or a bug in the algorithm presented
in the appropriate RFC?

Isn't UDP more suitable for data transfers?

-- 
Karel Kulhavy                     http://atrey.karlin.mff.cuni.cz/~clock

Re: TCP stack misbehaviour?

[kuznet]

Hello!

> empty, except for occasional ACKs. The utilization of the channel is about 4%.

1. tcpdump is required.
2. exact vesion of used kernel is required too.

Alexey

Re: TCP stack misbehaviour?

[clock]

On Sun, Apr 08, 2001 at 06:07:55PM +0200, Manfred Spraul wrote:
> Could you add more details and post them to linux-kernel?
> 
> * which kernel version

2.2.18

> * both sides linux, or one side linux, the other one something else

both sides 2.2.18 linux

> * a tcpdump that shows the problem, preferably 2 dumps, from both sides
> of the cable
> 
> --
>     Manfred
> 

-- 
Karel Kulhavy                     http://atrey.karlin.mff.cuni.cz/~clock