Jack Taylor writes:
> The *real solution* to this situation is to upgrade the quality/capacity of
> the RF links.
I feel this is not quite true. Observing the RF links shows the
*chaos* on the network and the congestion is in most cases at the
user uplink/downlink to/from the next/last node (observed at my site).
This is because of the collisions and the random transmission times
of multiple hosts and in case of simplex uplinks the impossibility to
hear other stations transmissions. This situation can be improved by
using DAMA and duplex uplinks in combination with some kind of
throughput control to not honour stations with aggressive parameters
and high data volume.
Traffic between nodes is mostly done with interlinks so there
are only 2 stations. This can be easily synchronized to minimize
collisions and maximize throughput. Normally excessive load will fill
up the interlink bandwidth but will not tend to break down the
overall throughput whereas excessive load on a user uplink tend to be
more inefficient the more load it carries.
The *real* solutions would be some kind of DAMA uplink with
the TCP/IP stack recognizing the state of the packets to send thus
setting timers accordingly to the uplink (i.e. starting timers
when the packet is really sent/acknowledged by the next
node/gateway). Measuring round trip times for uplink,
interlink and downlink can improve throughput if the measured values
are exchanged with the TCP/IP stack and affect the internal timers.
Just my $0.02
Bernd Laengerich
--
// c.a.r.u.s. Computer Service GmbH POS EFT Systeme //
// Borsteler Chaussee 49 Phone: +49-40-51435-0 //
// D-22453 Hamburg Fax: +49-40-51435-111 //
