[netperf-talk] Bug in UDP_STREAM test when built w/ omni tests enabled

Alexander Duyck alexander.h.duyck at intel.com
Thu Dec 1 14:57:57 PST 2011


On 12/01/2011 01:56 PM, Rick Jones wrote:
> On 12/01/2011 01:47 PM, Alexander Duyck wrote:
>> I pulled from top-of-trunk and it looks like it fixed it for me as well.
>
> Good - keep in mind that netperf has never supported mixing versions. 
> Often it will "just work" but that is "by chance" and there was a 
> change in the control message size post 2.5.0...
>
> happy benchmarking,
>
> rick jones
I saw that so I just went ahead and updated everything to the new 
version.  I also tried out the new "omni" test options and it looks like 
I found a few minor bugs.

The first one is on my x86_64 based system I am seeing:
OMNI %s TEST from ::0 (::) port 0 AF_INET6 to localhost (::1) port 0 
AF_INET6

instead of:
OMNI Send|Recv TEST from ::0 (::) port 0 AF_INET6 to localhost (::1) 
port 0 AF_INET6

The quick fix for that seemed to be to make direction_str in 
src/nettest_omni.c static.  Once I did that it fixed the issue.

The other thing I noticed is that Local CPU Util was always reporting as 
0%.  It turns out that was essentially the same issue, and once I made 
the variable declaration for local_cpu_utilization_double static it 
began reporting the correct value.  Odds are most of the global 
variables should probably be made static to avoid these kind of issues 
from cropping up.

Thanks,

Alex





More information about the netperf-talk mailing list