[netperf-talk] Some question about TCP_RR test

[Rick Jones]

On 07/22/2011 02:47 AM, Hangbin Liu wrote:
> On Thu, 2011-07-21 at 10:48 -0700, Rick Jones wrote:
>> On 07/21/2011 03:20 AM, Hangbin Liu wrote:
>>> Hi all,
>>>
>>>      I got a network traffic flood in our lab yesterday when I use TCP_RR.
>>> Could you help me answer the questions and find the flood reason?
>>
>> Traffic flood? With netperf TCP_RR?  FWIW, nothing that follows in the
>> email suggests (at least on the surface) a netperf-induced traffic flood
>> on your network.
>>
> Hi Rick,
>
>     Thank you very much for your response, I can get the detailed
> description form the url you give me. But there is actually a traffic
> flood in our lab. Our administrator said the flood was observed on
> another system, which *wasn't* being used as part of my testing on the
> lab network. He also gave me the tail end of tcpdump capture, which have
> been attached.
>     With the time in the attachment, I find the commond from my log.
>
> "netperf -6 -l 60 -H 2620:52:0:102f:20e:1eff:fe04:8190 -t TCP_STREAM -i
> 10,2 -I 99,5 -- -m 4096 -s 57344 -S 57344"
>
> But this is strange, ipv6 have no broadcast address, how can a ipv6 TCP
> test cause a traffic flood?? I have asked for more information about it
> and will tell you ASAP.


You should make sure that unicast MAC addresses are being used, and then 
see if perhaps the forwarding tables of your switch(es) are overflowing. 
  When that happens, switches have to flood traffic.

Neither of those are actually netperf issues though.

BTW, unless you are running a truly ancient version of netperf, the 
buffers will be filled with a repeating pattern of "netperf" so if you 
have tcpdump display the packet contents, that will further verify that 
those packets are netperf packets.

rick jones