[Bismark-devel] some 6to4 testing

[Dave Taht <d@taht.net>]

I've had a chance to do some basic testing of the most recent build of 
bismark at eric's place.

The good news: Most everything works great.

So, after getting 6to4 to work

(see: http://www.bufferbloat.net/issues/80 )

And having a few ongoing issues with multicast (still - see bug #81)

I did a bit of performance testing using iperf, over the wired interface.

I was seeing about 20% difference in speed on these two tests over 
comcast, between the 2002 and 2001 networks. Over FIOS... well....


IPv4 TEST:

d@cruithne:~$ iperf -t 60 -c huchra.bufferbloat.net
------------------------------------------------------------
Client connecting to huchra.bufferbloat.net, TCP port 5001
TCP window size: 16.0 KByte (default)
------------------------------------------------------------
[  3] local 192.168.1.181 port 56963 connected with 149.20.54.237 port 5001
[  3]  0.0-60.0 sec    176 MBytes  24.5 Mbits/sec
d@cruithne:~$ iperf -t 60 -V -c huchra.bufferbloat.net
------------------------------------------------------------
Client connecting to huchra.bufferbloat.net, TCP port 5001
TCP window size: 16.0 KByte (default)
------------------------------------------------------------
[  3] local 2002:47a2:f305:1::2 port 34130 connected with 
2001:4f8:3:36:2e0:81ff:fe49:2cfe port 5001
[  3]  0.0-63.6 sec  3.48 MBytes    459 Kbits/sec

It's about 165 ms to huchra via the nearest 6to4 router on FIOS (vs 
about 85ms on ipv4). BUT WOW, what a hit for "native <-> 6in4"!  Factor 
of 53 difference.

Going pure 6in4 replicates the result ratio I was getting with comcast:

d@cruithne:~$ iperf -V -t 60 -c 2002:9514:3640:36:2e0:81ff:fe23:90d3
------------------------------------------------------------
Client connecting to 2002:9514:3640:36:2e0:81ff:fe23:90d3, TCP port 5001
TCP window size: 16.0 KByte (default)
------------------------------------------------------------
[  3] local 2002:47a2:f305:1::2 port 37667 connected with 
2002:9514:3640:36:2e0:81ff:fe23:90d3 port 5001
^[[B[  3]  0.0-60.2 sec    160 MBytes  22.3 Mbits/sec


Which basically accounts for the size of the ack packets being larger 
and leveraging the existing ipv4 routing scheme.