[netperf-talk] Second line in output for DLCL_RR

[Rick Jones]

Alexander Vinokur wrote:
> Hi netperf,
>
>>  uname -a
> 
> HP-UX hpbl817 B.11.23 U ia64 0583057937 unlimited-user license
> 
> Netperf-2.4.4
>  
> Second line (netserver) always contains -1 -1.
> 
>  
> 
> I checked it also with dlpi0, dlpi1, dlpi2, dlpi3, dlpi4 for remote DLPI 
> device file name.
> 
> The same result.
>
> Any suggestions?
> 
> Thanks.
> 
> Alex Vinokur
> 
> ------------------------------------------------------
> Testing with the following command line:
> ./beta_netperf.hpbl817 -H hpbl818 -t DLCL_RR -- -D /dev/dlpi,/dev/dlpi 
> -p 0,0 -w 110,120 -W 70,80 -r 150,160
> 
> DLPI CL REQUEST/RESPONSE TEST
> Local /Remote
> Window Size   Request  Resp.   Elapsed  Trans.
> Send   Recv   Size     Size    Time     Rate        
> frames frames bytes    bytes   secs.    per sec  
> 
> 110    120    150      160     10.01    14117.17  
> -1     -1   

In a manner similar to a UDP test, those are the remote window settings, 
and for some reason, if my perusal of the code in src/nettest_dlpi.c is 
correct netserver is returning -1's for those values.

Looking further at the code, it seems that unless 
DL_HP_SET_LOCAL_WIN_REQ is set, they will end-up set to -1.

Based on:

$ find /usr/include -exec grep -l  DL_HP_SET_LOCAL_WIN_REQ {} \;

on an 11.11 system at my disposal, that define will come from 
/usr/include/sys/dlpi_ext.h

Going back to the top of the src/nettest_dlpi.h file it appears that it 
is included only when "__hpux__" is defined.  Perhaps that is not being 
defined by the compiler.

So, a couple of options:

*) edit your copy of the source to always include sys/dlpi_ext.h - this 
cannot happen in mainline because there is no guarantee that another 
platform with DLPI will have that include file

*) enhance the configure script source - configure.ac - to check for 
sys/dlpi_ext.h and replace the #ifdef __hpux__ with #ifdef 
HAVE_SYS_DLPI_EXT_H

happy benchmarking,

rick jones