[Olsr-dev] Test methodology input for BattleMesh

[Teco Boot]

Op 6 mrt 2010, om 22:03 heeft ZioPRoTo (Saverio Proto) het volgende geschreven:

>> I have packet delivery ratio of 98%.
>> I reduced Tx power recently to 10dBm.
>> With 23dBm, I had 99,8% and almost full connectivity.
>> Now, about half of nodes has link.
> 
> Teco, can you describe how you estimate the packet delivery ratio ? Do
> you use some tool like iperf to send and count received packets ? Do
> you have some scripts to perform the tests ?

I've 30 alix boxes with R52H, with both ethernets to high-speed VLAN infra.
There, two high-end iMacs run 30 XP hosts, on eth0 HNA. Alix monitoring
uses eth1, but no routes via this interface.

On XP hosts, NTP is running. NTP provides 8-bit shift register for reply
status (the packet delivery ratio), but also nice delay and jitter
measurement. Polling is fixed, 64 seconds. In total we have 30 * 29 * 8
"tests" for packet delivery ratio. Convergence takes 10 minutes (9 * 64).
We count get_time and reply_time as one, we are not marketeers :-) But 
for comparison, the loss should be halved. 
We wrote a multicast sender & receiver, with the 8-bit shift register. 
Reporting is similar to NTP, but no delay / jitter measurement.

Another XP host gets all info, logs and provides a nice dashboard.

Because the p2pd plugin (and mdns) runs on olsr messages, the multicast 
packet delivery ratio measurement is also useful for knowing loss on 
TC flooding. When Tx power was set to low value (0dBm), I saw some 
black holes in multicast delivery. I think the link cost threshold 
to low: when 1 out of 10 packets is delivered, TC loss is very likely.
When time, I take a look at this.

Regards, Teco