In our attempt to find out why we receive such a bad performance from our RF we have configured a linux box with 2 NICs to site between two machines connected via ethernet.
The linux box called canopy simulator runs a piece of software which makes changes to the traffic between the two machines.
straight transfer between two machines, a constant 80Mbs,
1ms delay, transfer drops to a constant 50Mbs
5ms delay, 10Mbs
20ms delay, 4Mbs (canopy have a 20ms delay)
in all the above tests the traffic monitoring graph showed a constant flow
the real interesing results were when we introduced packet drops.
no delay with 1% drops, speed a very spiky graph fluctuating between 20-80, average 50Mbs
1 ms 2% drops, very spiky 10Mbs
conclusion we were able to replicate teh graph to match what we see in Canopy by introducing delay and more importantly packet drops, hence Canopy does not handle packet drops too well…
The next stage of testing is to see if I can find out how many packets are being dropped and where, and then see if we can develop software to sit on either side to deal with this issue.
Any ideas welcome.
What software were you running this on?
The simulation software was running on a linux kernal, using NETEM, this is part of the linux kernal. We had to create the bridge between the two network cards. If you need the exact build and config files let me know and we will give you the exact build of the software.
We are now looking at the snoop protocol to see how we can better the performance.
All our canopy is on 7.2.9, we are also looking into the arp broadcast problem, it definately seems to be a bug issue with this software.