[HECnet] UDP tunnelling over SSH - Solution for people dynamic IP addresses?

[Sampsa Laine]

What I was referring to is the cascading time-out problems you can sometimes get tunnelling one TCP connection over another, e.g. HTTP over a PPP over SSH connection:

	3.	HTTP (TCP)
	2.	IP
	1.	PPP
	0. 	SSH (TCP)

Sometimes if a time-out occurs and layers 0 and 3 have different retransmission timers, the retransmissions kill the whole shebang.

More details can be found at: http://sites.inka.de/~W1011/devel/tcp-tcp.html

Not sure if this will be a problem with NSP over TCP however, but I can imagine it might be (if the NSP retransmission timer is faster than the underlying TCP timers).

Sampsa


On 13 May 2009, at 15:25, Paul Koning wrote:

"Sampsa" == Sampsa Laine <sampsa at mac.com> writes:

Sampsa> The main disadvantage I can see is that SSH runs over TCP so
Sampsa> any dropped packets might cause more delays than using
Sampsa> straight UDP.

That isn't actually a disadvantage when you travel all the way to the
top of the stack.   Yes, the DECnet network layer (just like IP) uses,
and provides, a datagram service.   But it also uses retransmit
internally for stuff that has to get through, and of course the
transport layer (NSP) makes a reliable service through timeout and
retry.

So a tunnel over TCP is just fine.   It means you have a lossless
network (ignoring congestion in the DECnet nodes).   So instead of
having delays due to timeout and retransmit in NSP, you have the same
delay (or, quite possibly, a shorter delay) due to timeout and
retransmit in TCP.   The overall application performance should come
out essentially the same.

    paul