[netperf-talk] Segfault report

[Rick Jones]

On 02/24/2012 02:23 AM, Gerd v. Egidy wrote:
> Hi,
>
> I just wanted to give a short report of a segfault I got from netperf when
> trying to output the THROUGHPUT_CONFID from an omni-test.
>
> Like this:
>
>> netperf -H 192.168.1.1,4 -I 99,5 -t omni -- -d stream -l 10 -O "PROTOCOL,LOCAL_BYTES_XFERD,THROUGHPUT,THROUGHPUT_UNITS,THROUGHPUT_CONFID" -m 64K
> Segmentation fault (core dumped)
>
> But this works as expected:
>
>> netperf -H 192.168.1.1,4 -I 99,5 -t omni -- -d stream -l 10 -O "PROTOCOL,LOCAL_BYTES_XFERD,THROUGHPUT,THROUGHPUT_UNITS" -m 64K
>
> I haven't done any further investigation as I'm in the middle of benchmarking ;)
>
> Could you try to reproduce it?

Well, in my quick test on a 64-bit Ubuntu 11.04 system using 
top-of-trunk I don't see it:

raj at tardy:~/netperf2_trunk/src$ ./netperf -H localhost -I 99,5 -t omni 
-- -d stream -l 10 -o 
"PROTOCOL,LOCAL_BYTES_XFERD,THROUGHPUT,THROUGHPUT_UNITS,THROUGHPUT_CONFID" 
-m 64K
establish control: are you sure there is a netserver listening on 
localhost at port 12865?
establish_control could not establish the control connection from 
0.0.0.0 port 0 address family AF_UNSPEC to localhost port 12865 address 
family AF_INET
raj at tardy:~/netperf2_trunk/src$ ./netserver
Starting netserver with host 'IN(6)ADDR_ANY' port '12865' and family 
AF_UNSPEC
raj at tardy:~/netperf2_trunk/src$ ./netperf -H localhost -I 99,5 -t omni 
-- -d stream -l 10 -O 
"PROTOCOL,LOCAL_BYTES_XFERD,THROUGHPUT,THROUGHPUT_UNITS,THROUGHPUT_CONFID" 
-m 64K
OMNI Send TEST from 0.0.0.0 (0.0.0.0) port 0 AF_INET to 
localhost.localdomain () port 0 AF_INET : +/-2.500% @ 99% conf.  : demo
Protocol Local        Throughput Throughput  Throughput
          Bytes                   Units       Confidence
          Xferred                             Width (%)

TCP      45847150592  36808.65   10^6bits/s  4.520

BTW, the "-l 10" is misplaced - that should be a global option.  Also, 
if I recall correctly, the local bytes transferred will not have gone 
through the averaging that the throughput will have when confidence 
intervals are requested, so there may be a discrepancy from time to time 
as the local bytes transferred will be for the last iteration only.

Oh, and a top-of-trunk test with a 32-bit Ubuntu 11.10 system seems to 
have completed just fine with the command line I've used above, and they 
both are fine when I remember to tack-on the ",4" after the hostname.

So, apart from the usual support song and dance of "Have you tried the 
latest bits" :) what is your environment?

happy benchmarking,

rick jones