noud,

You asked about going faster than 57600. We can't do that yet. Most
machines don't support it, and we have no documentation for the
ones which do. We'll need to have DMA support for serials first.

> >Hello,
> >On Wed, 20 Mar 1996, noud de brouwer wrote:
> >
> >>
> >> Recently i enlarged the ring buffer i use,
> >> this gives a speed improvement from 3.1k/s to 3.4-3.9k/s throughput.
> >>
> >I'm using 57.6 Kb/s over a null-modem cable a ppp link to my Linux-Box and
> >there I get between 5 KB/s and 6 KB/s (never below 5KB/s).
> 
> Yes, on the machine doing ppp i get +/-5k/s too.
> But..that one is the gateway for the rest of the machines, done by ether.
> Now when i'm on one of the other machines (not the gateway) the total
> 'throughput' of the 'signal' drops to 3.1k/s, that is 3.6k/s nowadays.
> 
> (yes i know..the machine is very bussy)
> 
> But when i do a transfer from the net to the gateway &
> a transfer from the gateway to another local machine..
> at the same time, that last transfer over ether still has >25k/s and the
> serial connections something around 5k/s.
> So where does it's time go??

You have:

  ISP <== 57600 modem ==> gateway <== Ether ==> computer

So you get ~ 5 K/s ISP <-> gateway, and ~ 25 K/s gateway <-> computer,
and ~ 3 K/s ISP <-> computer. (or something like that; basically you
go fast over each segment, and wonder why it slows down when you
straddle segments)

Sounds about right. The problem is that the packet has to FULLY go
through the serial link, get processed, then FULLY go over the
ether link. So basically the times required for each step add.

1/Rate = 1/serial_rate + 1/ether_rate + overhead

Rate = 1/(1/5K + 1/25K + overhead)    If we set overhead = 2/25K (just
                                            picking a number at random)

Rate = 1/(1/5 + 1/25 + 2/25) = 3.1 K/s

So your numbers make sense.

Take care,

Bill