[netperf-talk] DCCP Unidirectional with Omni Tests
Tim Schneider
schneid5 at cs.uni-bonn.de
Thu Aug 12 10:56:26 PDT 2010
Hello,
I actually stumbled on a way to fix it, though I haven't put any more thought into why that actually fixes it.
I ran the following three commands and had no errors since:
sudo sysctl -w net.dccp.default.seq_window=10000
sudo sysctl -w net.dccp.default.rx_ccid=3
sudo sysctl -w net.dccp.default.tx_ccid=3
don't know if switching to CCID3 did the trick, or maybe of the seq_window was just to small...
Maybe this helps someone in the future.
Greez,
Tim Schneider
Am 12.08.2010 um 18:46 schrieb Rick Jones:
> Tim Schneider wrote:
>> Hey everyone,
>> I got Netperf to compile since my last mail, but now I stumbled onto the next
>> Problem.
>> I am trying to use Netperf to measure unidirectional DCCP Traffic. I am using
>> Omni tests, since there's no other test, that allows to use DCCP (correct?).
>
> Yes, there is no src/nettest_dccp.c and corresponding DCCP_mumble tests.
>
>> When I start the test, Netperf starts to send DCCP Packets, but after
>> successfully sending a couple of them, i get a message
>> "netperf: send_omni: send_data failed: Resource temporarily unavailable
>> send_data: data send error: errno 1"
>> I'm calling netperf with these commandline options:
>> netperf -t omni -H 127.0.0.1 -- -T dccp -m 1464 (localhost is just for
>> testing the software, I will switch to a real network afterwards.)
>> when I add -r 1464, the test works just fine. Since I don't want to test
>> Request/Response, this is no solution for me.
>> Omni tests also work fine with TCP, UDP and SCTP.
>> It would be great if someone could help me with this.
>
> You are likely the first person to do much of anything "meaningful" with DCCP in the omni tests.
>
> Presently, netperf expects to get "errors" only on a UDP send call, so getting one while sending DCCP will cause it to abort. The question becomes how "normal" the error it is getting with DCCP happens to be and if then it needs to be special cased.
>
> happy benchmarking,
>
> rick jones
More information about the netperf-talk
mailing list