[netperf-talk] A second tentative migration step has been taken.

Rick Jones rick.jones2 at hp.com
Tue Jul 20 16:59:36 PDT 2010


Folks -

I have taken a second, tentative step towards migrating the myriad of TCP/UDP 
tests in src/nettest_bsd.c to the "omni" code in src/nettest_omni.c.  That step 
consists of some slight tweaks to the omni code itself, along with the addition 
of the shell of the send_tcp_stream() (what netperf executes for a TCP_STREAM 
test) code to src/nettest_omni.c.  This "shell game" is controlled by an ifdef 
called "WANT_MIGRATION" which is not presently controlled via the configure script.

If you ./configure --enable-omni  and then compile, all will be as it has been - 
omni tests and TCP_STREAM separate.  If though you edit config.h to #define 
WANT_MIGRATION 1 and recompile, the TCP_STREAM test will not run the routine in 
src/nettest_bsd.c but the routine in src/nettest_omni.c.  This shell in 
src/nettest_omni.c is just the original output code, which preserves the 
formatting for "legacy" tests.  There is one small output difference - for now 
the migrated test will say "Migrated TCP STREAM" rather than TCP STREAM.

One other difference is that with WANT_MIGRATION set, scan_omni_args() will be 
called rather than scan_socket_args() - in theory, scan_omni_args is a superset 
of scan_socket_args().

If you use any "omni unique" output format options (-o, -O or -k), the legacy 
output will be bypassed and the omni output routines will be used just as if one 
had used netperf -t omni instead of netperf -t tcp_stream

The plan is that the legacy netperf test names for TCP and UDP (not sure about 
SCTP) will simply set some initial values for the omni tests, which can then be 
overridden.

Any and all feedback is encouraged.

happy benchmarking,

rick jones


More information about the netperf-talk mailing list