BE SURE TO READ THE MANUAL. EVEN THOUGH IT MAY BE OUTDATED.

This is a brief readme file for the netperf TCP/UDP/sockets/etc
performance benchmark. This is here mostly as a boot-strap. The real
information is in the manual, which can be found in netperf.ps and
online from http://www.netperf.org/netperf/NetperfPage.html. The
sources, and a limited number of binaries, can be found from
ftp://ftp.cup.hp.com/dist/networking/benchmarks/netperf/ .

BE SURE TO READ THE MANUAL. EVEN THOUGH IT MAY BE OUTDATED.

There is a COPYRIGHT file included.  It is called COPYING because that
is what autosomethingorother wanted. It is based on what the HP Legal
Eagles gave me. I am not sure if the legalese is clear, but the intent
is to say "Here is a benchmark. Use it in good health. Pass it along,
port it, enhance it. You didn't pay for this tool, so don't expect it
to be perfect ;-)" The rest of it is there to keep the layers happy...

While the copyright is pretty much in spirit an "open source" one, it
is not in letter - I never took the time to try to get it approved.
If you feel strongly about the license of the code you use and want
something under the GPL, consider netperf4:

http://www.netperf.org/svn/netperf4/trunk

Feel free to report netperf results in public forums, but please be
excruciatingly complete in your description of the test envorinment.
The old netperf database at:

  http://www.netperf.org/netperf/NetperfPage.html

is no more - or rather the utilities for accessing it no longer run.
The data is still present in the tree, albeit _VERY_ old now.

There is an Internet mailing list devoted to netperf. It is called
netperf-talk and it is hosted on netperf.org. Subscription requests
should go to netperf-talk-request@netperf.org.

Please DO NOT SEND subscription requests to netperf-talk!

If you run into severe difficulties, or are just feeling chatty,
please feel free to drop some email to me - Rick Jones
<rick.jones2@hp.com>. Be sure to include a meaningful subject lines.

happy benchmarking,
rick jones

BE SURE TO READ THE MANUAL. EVEN THOUGH IT MAY BE OUTDATED.

Pseudo-random things about netperf on AIX:

While it _should_ not be necessary in the release bits, the rc bits
for netperf 2.4.0 required:

    configure _may_ need:

    CFLAGS="-qcpluscmt -lperfstat"

    until such time as Rick Jones can figure-out or be told how to
    automagically add those using the configure script (hint hint :)

The release bits should be OK without the above.  Depending on the
name used to invoke the compiler, the -qmumble option may be implicit.

AIX include files have a VERY unfortuneate set of #define's in them
for phrases network oriented programs are QUITE likely to have in
their source - "rem_addr" and "rem_size" A "cousin" of the Netperf
Contributing Editor reports this interferes with --enable-dlpi
compilation and that it was also a problem for MySQL compiltion.
While we await IBM's APAR with bated breath, netperf has been kludged
to workaround this bug in IBM's include files.  It has been reported
that a "PMR" 34940,212,848 has been submitted to IBM in relation to
this bug in their header files.

Coutesy of Jan Krueger of TUI Infotec, Germany we have a pointer to
a workaround for problems with compilation on AIX 6.1:

https://www-304.ibm.com/support/docview.wss?uid=isg1IV01736

when you get errors like:
config.status: creating src/netperf_version.h
        gcc -DHAVE_CONFIG_H -I. -I..       -MT netlib.o -MD -MP -MF .deps/netlib.Tpo -c -o netlib.o netlib.c
In file included from /usr/include/sys/corral.h:25,
                 from /usr/include/libperfstat.h:28,
                 from netlib.h:679,
                 from netlib.c:166:
/usr/include/netinet/in6_var.h:65: error: array type has incomplete element type

A note about CPU utilization...

For HP-UX 11.0 <= system < 11.23 the configure script will select the
"pstat" CPU utilization mechanism.  This mechanism is the familiar
HP-UX idle counter mechanism (for all incense and porpoises) and
requires calibration.  See src/netcpu_pstat.c for all the details.

For HP-UX 11.23 >= system, the configure script will select the
"pstatnew" CPU utilization mechanism.  11.23 adds cycle counts for
user, kernel and interrupt modes to the idle cycle counter.  As such,
it _should_ be possible to simply take the sum of the four and the
fractions and know how much time was spent in each mode.
HOWEVER... there is a bug in the accounting for interrupt cycles,
where interrupt cycles go missing.  SOOO, since there is an accurate
way to know what the total number of cycles should have been over the
interval, and we know (ass-u-me) that the idle cycle counter is good
(since the pstat mechanism has tested that one OK), we will take the
ratio of idle to total cycles to compute CPU util.

We will still calculate fractions for user, kernel and interrupt, and
report them in debug (-d) output, but with a warning for interrupt
time.  See src/netcpu_pstatnew.c for all the details.

Up through HP-UX 11.23 (aka 11iV2) if you enable burst mode, and
happen to send sub-MSS requests and/or responses you _cannot_ assume
that the packet per second rate on the wire will match the transaction
per second rate reported by netperf, even if you set TCP_NODELAY with
the test-specific -D option.  The HP-UX 11.X TCP stack likely will be
generating some immediate 'standalone' ACKnowledgements which may not
be generated by other stacks.  This has been reported to the HP-UX TCP
folks, and an announcement will be made when that issue is resolved.

If you are reading this, it suggests you are using a version of
netperf in which the issue of file/directory name case insensitivity
in OSX has been worked-around by renaming the NetPerf and NetServer
directories to NetPerfDir and NetServerDir respectively.

February 11, 2003

At the time of the initial port, I was not aware of a make facility
for OpenVMS.  So, I would just compile and link the various files by
hand:

   $ cc netperf.c
   $ cc netlib.c
   $ cc netsh.c
   $ cc nettest_bsd.c
   $ cc netserver.c
   $ link/exe=netperf netperf.obj,netsh.obj,netlib.obj,nettest_bsd.obj
   $ link/exe=netserver netserver.obj,netsh.obj,netlib.obj,nettest_bsd.obj

Installation for OpenVMS has a few differences from installation under
say Unix.  There is no inetd for VMS - however, there is the concept
of an adding an auxilliary service that seems quite similar.

To configure netperf for operation as an auxilliary service, you will
need to edit/use the netserver_run.com file and alter the "path" to
netserver accordingly.  The version that ships is setup for where Rick
Jones did his initial porting work and most likely is not apropriate
for you :)

   $  define sys$output sys$sysroot:[netperf]hello_service.log
   $  define sys$error  sys$sysroot:[netperf]hello_service.log
   $  run sys$sysroot:[netperf]netserver.exe

Then it will be necessary to "define" netperf (netserver) as an
auxilliary service. This will need to be customized as apropriate for
your system

   $ tcpip set service netserver -
   _$ /port=12865 -
   _$ /protocol=tcp -
   _$ /user=system -
   _$ /limit=48 -
   _$ /process_name=netserver -
   _$ /file=sys$sysroot:[netperf]netserver_run.com

And then it is necessary to enable the service:

$ tcpip enable service netserver

If you want to disable the service, you can issue the command

$ tcpip set noservice netserver

By default, OpenVMS is case-insensitive with commandlines, and will
downshift everything to lower case.  This does not interact well with
netperf's use of command-line options like "-H" and "-h" with rather
different meanings.

To workaround that, the following defines are believed to be
sufficient.

$ define  DECC$ARGV_PARSE_STYLE       ENABLE
$ define  DECC$EFS_CASE_PRESERVE      ENABLE
$ define  DECC$POSIX_SEEK_STREAM_FILE ENABLE
$ define  DECC$EFS_CHARSET            ENABLE
$ set process /parse_style=extended

I do not know if this will be something one can also do for the
netserver - presumeably one could put these things in the
netserver_run.com file (guessing). At present though I've not tried
that, and I'm also not sure that netserver has any upper-case options.

Until the release bits the following was true:

    Until such time as Rick Jones can figure-out or be told how to make it
    automagic in the configure script, prior to configure on solaris, you
    may need:

    CFLAGS="-lsocket -lnsl -lkstat"

    and if you are trying to compile the SCTP tests:

    CFLAGS="-lxnet -lsocket -lnsl -lkstat -D_XOPEN_SOURCE=500 -D__EXTENSIONS__"

as the release bits have a "smarter" configure script, and the SCTP
tests use libsctp, the above no longer applies.  It should all just be
automagic (although for SCTP you still must --enable-sctp at configure
time)

Beware CPU util figures on anything before Solaris 10 that does not
say 100%, and still be a triffle cautious with Solaris 10 CPU util
reports.  The CPU time accounting mechanisms either do not track time
spent servicing interrupts, or do so in parallel with time spent in
user/kernel/idle which means that some idle time isn't _really_ idle
time.

And beyond that, it is still not clear if the CPU utilization reported
on systems with hardware threading support (eg UltraSPARC-T1) is
really accurate even ignoring the issue with interrupt time.  It is
likely that to be truely accurate, it is necessary to know how much
"real work" any one strand performed.

Compiling for VMware is somewhat like compiling for Windows - there is
a separate, standalone makefile one uses.  In this case, it is
src/Makefile.uw.  So, to build the bits, cd to src/ and make -f
Makefile.uw.

At present, the makefile is setup to use a number of the "none" files
- in particular netcpu_none.c.  When/if we enable the "omni" tests
we'll perhaps see the addition of a number of other "none" files as
well.

Also, seems the way things are "run" under VMware is enough different
that the scripts, should you chose to use them, will need to be
modified.  The initial set of patches make some arbitrary changes that
need to be re-worked with some "To run this under VMware uncomment
this line" or somesuch.

It has been reported that versions of netperf have configured and
compiled under Cygwin.

It is also known that netperf has compiled using the Windows DDK.
Here is a skeleton of the instructions to do so:

Steps are:

A) Install the Windows driver developer kit (if not already done).

B) open a Cmd windows (i.e., a DOS box) for the target environment
   (target OS version; free vs checked build; x86, AMD64, or IA64).
   This is picked from the "Start\Developer Kits" path.

C) enter the src\NetPerfDir directory

D) Edit sources to enable any desired optional features (eg
   -DWANT_HISTOGRAM) or to remove features which your version of
   Windows might not support (eg -DHAVE_STRUCT_SOCKADDR_STORAGE)

E) while still in the src\NetperfDir directory type "build /cD".

F) Repeat steps C through E in src\NetServerDir

G) the target files will be in a directory like:
   NetPerfDir\objchk_wnet_IA64\IA64, NetServerDir\objchk_wnet_IA64\IA64
   NetPerfDir\objfre_wnet_x86\i386, or NetPerfDir\objfre_wnet_AMD64\amd64

NOTE: If any components of the path (ie the full names of the files,
including parent directories) contain spaces (eg "My Documents"),
build will charge off into the weeds.

If you do not want the corresponding features, edit the sources files
and remove the -Ds for WANT_DEMO, WANT_HISTOGRAM and/or WANT_INTERVALS

And if that weren't enough, it is also known that netperf has been
compiled using MS Visual Studio 2003.  Here are the instructions from
the person who made that work (See Authors):

1.  Under the PROJECT tab, <project name> PROPERTIES, LINKER folder,
Select COMMAND LINE and add WS2_32.lib in the whitespace labeled
Additional Options:


2.  Under the PROJECT tab, <project name> PROPERTIES, C/C++ foleder,
Select Preprocessor, On the right, add DO_IPV6; at the end of the
Preprocessor Definitions whitespace.


He goes on to say:

NOTE: WHEN COMPLING NETSERVER, it works, but I got issued the
foillowing warnigns in my build:


------ Rebuild All started: Project: netserver, Configuration: Debug Win32 ------

Deleting intermediate files and output files for project 'netserver', configuration 'Debug|Win32'.

Compiling...

nettest_bsd.c

g:\Program Files\netperf\netperf-2.4.1rc1\src\nettest_bsd.c(846) : warning C4267: 'function' : conversion from 'size_t' to 'int', possible loss of data

g:\Program Files\netperf\netperf-2.4.1rc1\src\nettest_bsd.c(1303) : warning C4267: 'function' : conversion from 'size_t' to 'int', possible loss of data

g:\Program Files\netperf\netperf-2.4.1rc1\src\nettest_bsd.c(2020) : warning C4267: 'function' : conversion from 'size_t' to 'int', possible loss of data

g:\Program Files\netperf\netperf-2.4.1rc1\src\nettest_bsd.c(5080) : warning C4267: 'function' : conversion from 'size_t' to 'int', possible loss of data

g:\Program Files\netperf\netperf-2.4.1rc1\src\nettest_bsd.c(5715) : warning C4267: 'function' : conversion from 'size_t' to 'int', possible loss of data

g:\Program Files\netperf\netperf-2.4.1rc1\src\nettest_bsd.c(6591) : warning C4267: 'function' : conversion from 'size_t' to 'int', possible loss of data

g:\Program Files\netperf\netperf-2.4.1rc1\src\nettest_bsd.c(8013) : warning C4267: 'function' : conversion from 'size_t' to 'int', possible loss of data

g:\Program Files\netperf\netperf-2.4.1rc1\src\nettest_bsd.c(11123) : warning C4267: 'function' : conversion from 'size_t' to 'int', possible loss of data

netsh.c

netserver.c

g:\Program Files\netperf\netperf-2.4.1rc1\src\netserver.c(457) : warning C4267: 'function' : conversion from 'size_t' to 'int', possible loss of data

netlib.c

g:\Program Files\netperf\netperf-2.4.1rc1\src\netlib.c(2470) : warning C4267: 'function' : conversion from 'size_t' to 'int', possible loss of data

g:\Program Files\netperf\netperf-2.4.1rc1\src\netlib.c(2480) : warning C4267: 'function' : conversion from 'size_t' to 'int', possible loss of data

netcpu_ntperf.c

inet_ntop.c

Generating Code...

Linking...

Build log was saved at "file://g:\Documents and Settings\Administrator\My Documents\Visual Studio Projects\netserver\Debug\BuildLog.htm"

netserver - 0 error(s), 11 warning(s)



---------------------- Done ----------------------

Rebuild All: 1 succeeded, 0 failed, 0 skipped