Hi,

if 2 machines are connected to an ethernet, a fast one with 100MBit (A),
the other with 10MBit (B) one can often see that data transfered from A to B
via TCP drops to about 100Kbit/s. If they are connected to the same switch
activating back pressure in half duplex (or flow control with full duplex) on
the 100MBit port "fixes" this.

The question is: why does TCP (which should do flow control) does handle the
situation so bad?

It seems that TCP sends to fast. Switch buffers get full. Switch starts
dropping packets. TCP needs to retransmit - and this slows down the whole
connection.

But shouldn't the end-end flow control of TCP should limit the connection
to about 10Mbit/s which again means the switch would not have to drop
packets.

Is there any projects to change TCPs flow control mechanisms to behave better?
Because the ethernet flow control / back pressure may be usefull if both
machines are on the same net. It seems to cause more harm then good if they
are on different ports (from my experiences).

Greetings

Wolfgang Walter
-
To unsubscribe from this list: send the line "unsubscribe linux-net" in
the body of a message to [EMAIL PROTECTED]

Reply via email to