Suspect the opinions on this topic depends on the persective it is viewed.
Assuming segments of multiple user netrom RF links are involved, an observer
will see considerable QRM generated by link to link netrom acks under VC
mode when the network is fairly active. Eventually this situation will
cause the netrom users to time out and when the congestion is cleared, the
VC traffic will pass. Of course the other users won't be happy, but the
TCPers (not seeing the chaos their traffic causes) will be pleased.
DG mode on the other hand creates much less QRM during congested situations,
but at the cost of delaying the traffic until the network is clear.
The *real solution* to this situation is to upgrade the quality/capacity of
the RF links.
73 de Jack
Julian Munoz Dominguez wrote:
> John Cusick wrote:
>
> > Essentially, what are the differences and why or when would I
> > choose one and/or the other?
>
> Hello John, as you see there is a variety of opinions !!
> My conclusions are:
>
> If you have a very good link with the gateway of your subnetwork, use
> datagram.
> If you have a bad link, use connected mode.
>
> If your link is a normal link, then feel free to do experiments :-)
>
> For me, the conclusions of my experiments is that the interaction of tcp
> with ax25 depends of the the IMPLEMENTATION of TCP, in concrete with the
> Initial retransmition timers.
>
> With Linux I am having problems in VC mode, because:
> - IRRT value given in the route command is only used before the ACK SYN
> is received (the connection is established). After it is reseted. So for
> example if I use a client like Netscape, the fisrt request is beeing
> retransmited very much !!! For example I can wait be 2 minutes or more
> retransmiting the 5 o 6 times the first "GET http://... [lot of stuff
> generated by Mozilla]." I don't know it that has been improved in 2.2
> series. This improvement is compatible with RFC.
> - TCP maximum retransmition is 2 minutes. So if your packets take more
> than 2 minutes to cross the network, tcp generates congestion. This
> behaviour of 2min is imposed by the RFC, but jnos didn't implement it,
> wich is better for us.
>
> With jnos VC worked very well. With Linux I am seeing that ax25 is not
> working well, this added to the Linux TCP features do that VC is not
> working well in Linux.
> But I still use it, because I want to track all the improvements in this
> field.
>
>
>
> --
> Saludos de Julian
> -.-
> >> Espacio libre para poner publicidad de RadioInternet <<
