[Codel] iperf3 udp flood behavior at higher rates

Dave Taht dave.taht at gmail.com
Mon May 2 17:07:23 PDT 2016


On Mon, May 2, 2016 at 4:27 PM, Rick Jones <rick.jones2 at hpe.com> wrote:
> On 05/02/2016 04:18 PM, Dave Taht wrote:
>>
>> Netperf does not have a multi-hop capable udp flood test (rick jones
>> can explain why... )
>
>
> Well, with an intro like that, how could I refuse?-)
>
> In a nutshell, after repeated uses of the netperf UDP_STREAM test in
> non-air-gapped test setups with link up/down testing and a default route
> which pointed at the local "corporate/desktop/whathaveyou" network by QA
> engineers who should have known better, taking-out things like security
> camera video feeds, by default the data socket for a UDP_STREAM test is set
> SO_DONTROUTE.  This can be overridden with a test-specific -R 1 option.

That is so much more PC than what's actually in the release notes! :)
Well, a quick recompile later and applying -R 1 still gets rejected...
reading nettest_omni.c was quite entertaining, tho.

I actually like a world where it takes more effort to mess up the
network - where ping -f is not available to anyone but root, for
example. It's bothersome, in flent, to need root to get a ping
--step-size resolution of < 20ms, but I can live with it, and I'd like
a world where script kiddies with "ion cannons" had their hardware
explode on them when mis-used...

There are some things ordinary man does not need to know, and
knowledge of the netperf -t UDP_STREAM -R 1 option is one of them.

Except, maybe today, when I'm trying to figure out why iperf3 is being weird.

> This is not necessary if the test is UDP_RR or TCP_*.
>
> happy benchmarking,
>
> rick jones



-- 
Dave Täht
Let's go make home routers and wifi faster! With better software!
http://blog.cerowrt.org



More information about the ath10k mailing list