xref: /iperf/docs/dev.rst (revision 84b24b89)
1iperf3 Development
2==================
3
4The iperf3 project is hosted on GitHub at:
5
6http://github.com/esnet/iperf
7
8This site includes the source code repository, issue tracker, and
9wiki.
10
11Mailing Lists
12-------------
13
14The developer list for iperf3 is:  [email protected].
15Information on joining the mailing list can be found at:
16
17http://groups.google.com/group/iperf-dev
18
19There is, at the moment, no mailing list for user questions, although
20a low volume of inquiries on the developer list is probably
21acceptable.  If necessary, a user-oriented mailing list might be
22created in the future.
23
24Bug Reports
25-----------
26
27Before submitting a bug report, try checking out the latest version of
28the code, and confirm that it's not already fixed.  Then submit to the
29iperf3 issue tracker on GitHub:
30
31https://github.com/esnet/iperf/issues
32
33**Note:** Issues submitted to the old iperf3 issue tracker on Google
34Code (or comments to existing issues on the Google Code issue tracker)
35will be ignored.
36
37Changes from iperf 2.x
38----------------------
39
40New options (not necessarily complete, please refer to the manual page
41for a complete list of iperf3 options)::
42
43    -V, --verbose             more detailed output than before
44    -J, --json                output in JSON format
45    -Z, --zerocopy            use a 'zero copy' sendfile() method of sending data
46    -O, --omit N              omit the first n seconds (to ignore slowstart)
47    -T, --title str           prefix every output line with this string
48    -F, --file name           xmit/recv the specified file
49    -A, --affinity n/n,m      set CPU affinity (Linux and FreeBSD only)
50    -k, --blockcount #[KMG]   number of blocks (packets) to transmit (instead
51                              of -t or -n)
52    -L, --flowlabel           set IPv6 flow label (Linux only)
53
54Changed flags::
55
56    -C, --linux-congestion    set congestion control algorithm (Linux only)
57                              (-Z in iperf2)
58
59
60Deprecated flags (currently no plans to support)::
61
62    -d, --dualtest           Do a bidirectional test simultaneously
63    -r, --tradeoff           Do a bidirectional test individually
64    -T, --ttl                time-to-live, for multicast (default 1)
65    -x, --reportexclude [CDMSV]   exclude C(connection) D(data) M(multicast)
66                                  S(settings) V(server) reports
67    -y, --reportstyle C      report as a Comma-Separated Values
68
69Also deprecated is the ability to set the options via environment
70variables.
71
72Known Issues
73------------
74
75The following problems are notable known issues, which are probably of
76interest to a large fraction of users or have high impact for some
77users, and for which issues have already been filed in the issue
78tracker.  These issues are either open (indicating no solution
79currently exists) or closed with the notation that no further attempts
80to solve the problem are currently being made:
81
82* UDP performance: Some problems have been noticed with iperf3 on the
83  ESnet 100G testbed at high UDP rates (above 10Gbps).  The symptom is
84  that on any particular run of iperf3 the receiver reports a loss
85  rate of about 20%, regardless of the ``-b`` option used on the client
86  side.  This problem appears not to be iperf3-specific, and may be
87  due to the placement of the iperf3 process on a CPU and its relation
88  to the inbound NIC.  In some cases this problem can be mitigated by
89  an appropriate use of the CPU affinity (``-A``) option.  (Issue #55)
90
91* Interval reports on high-loss networks: The way iperf3 is currently
92  implemented, the sender write command will block until the entire
93  block has been written. This means that it might take several
94  seconds to send a full block if the network has high loss, and the
95  interval reports will have widely varying interval times.  A
96  solution is being discussed, but in the meantime a work around is to
97  try using a small block size, for example ``-l 4K``.  (Issue #125,
98  a fix will be released in iperf 3.1)
99
100* The ``-Z`` flag sometimes causes the iperf3 client to hang on OSX.
101  (Issue #129)
102
103* On OpenBSD, the server seems to require a ``-4`` argument, implying
104  that it can only be used with IPv4.  (Issue #108)
105
106* When specifying the TCP buffer size using the ``-w`` flag on Linux,
107  the Linux kernel automatically doubles the value passed in to
108  compensate for overheads.  (This can be observed by using
109  iperf3's ``--debug`` flag.)  However, CWND does not actually ramp up
110  to the doubled value, but only to about 75% of the doubled
111  value.  Some part of this behavior is documented in the tcp(7)
112  manual page.  (Issue #145)
113
114* On some platforms (observed on at least one version of Ubuntu
115  Linux), it might be necessary to invoke ``ldconfig`` manually after
116  doing a ``make install`` before the ``iperf3`` executable can find
117  its shared library.  (Issue #153)
118
119There are, of course, many other open and closed issues in the issue
120tracker.
121
122Versioning
123----------
124
125iperf3 version numbers use (roughly) a `Semantic Versioning
126<http://semver.org/>`_ scheme, in which version numbers consist of
127three parts:  *MAJOR.MINOR.PATCH*
128
129The developers increment the:
130
131* *MAJOR* version when making incompatible API changes,
132
133* *MINOR* version when adding functionality in a backwards-compatible manner, and
134
135* *PATCH* version when making backwards-compatible bug fixes.
136
137Release Engineering Checklist
138-----------------------------
139
1401. Update the ``README`` and ``RELEASE_NOTES`` files to be accurate. Make sure
141   that the "Known Issues" section of the ``README`` file is up to date.
142
1432. Compose a release announcement.  Most of the release announcement
144   can be written before tagging.  Usually the previous version's
145   announcement can be used as a starting point.
146
1473. Preferably starting from a clean source tree (be sure that ``git
148   status`` emits no output)::
149
150    vi RELEASE_NOTES   # update version number and release date
151    vi configure.ac    # update version parameter in AC_INIT
152    vi src/iperf3.1    # update manpage revision date if needed
153    vi src/libiperf.3  # update manpage revision date if needed
154    git commit -a      # commit changes to the local repository only
155    ./bootstrap.sh     # regenerate configure script, etc.
156    git commit -a      # commit changes to the local repository only
157
158    # Assuming that $VERSION is the version number to be released...
159    ./make_release tag $VERSION # this creates a tag in the local repo
160    ./make_release tar $VERSION # create tarball and compute SHA256 hash
161
162    # Testing of the release artifact happens here.  When all looks
163    # satisfactory (but not before that), then...
164    git push            # Push version changes
165    git push --tags     # Push the new tag to the GitHub repo
166
167   These steps should be done on a platform with a relatively recent
168   version of autotools / libtools.  Examples are MacOS / MacPorts or
169   FreeBSD.  The versions of these tools in CentOS 6 are somewhat
170   older and probably should be avoided.
171
1724. Stage the tarball (and a file containing the SHA256 hash) to the
173   download site.  Currently this is located on ``downloads.es.net``.
174
1755. From another host, test the link in the release announcement by
176   downloading a fresh copy of the file and verifying the SHA256
177   checksum.  Checking all other links in the release announcement is
178   strongly recommended as well.
179
1806. Also verify (with file(1)) that the tarball is actually a gzipped
181   tarball.
182
1837. For extra points, actually try downloading, compiling, and
184   smoke-testing the results of the tarball on all supported
185   platforms.
186
1878. Plug the SHA256 checksum into the release announcement.
188
1899. PGP-sign the release announcement text using ``pgp --clearsign``.
190   The signed announcement will be sent out in a subsequent emails,
191   but could also be archived.  Decoupling the signing from emailing
192   allows a signed release announcement to be resent via email or sent
193   by other, non-email means.
194
19510. Send the PGP-signed release announcement to the following
196    addresses.  Remember to turn off signing in the MUA, if applicable.
197
198    * [email protected]
199
200    * [email protected]
201
202    * [email protected]
203
204    * [email protected]
205
206    Note: Thunderbird sometimes mangles the PGP-signed release
207    announcement so that it does not verify correctly.  This could be
208    due to Thunderbird trying to wrap the length of extremely long
209    lines (such as the SHA256 hash).  Apple Mail and mutt seem to
210    handle this situation correctly.  Testing the release announcement
211    sending process by sending a copy to oneself first and attempting
212    to verify the signature is highly encouraged.
213
21411. Update the iperf3 Project News section of the documentation site
215    to announce the new release (see ``docs/news.rst`` and
216    ``docs/conf.py`` in the source tree) and deploy a new build of the
217    documentation to GitHub Pages.
218
219Code Authors
220------------
221
222The main authors of iperf3 are (in alphabetical order):  Jon Dugan,
223Seth Elliott, Bruce A. Mah, Jeff Poskanzer, Kaustubh Prabhu.
224Additional code contributions have come from (also in alphabetical
225order):  Mark Ashley, Aaron Brown, Aeneas Jaißle, Susant Sahani, Brian
226Tierney.
227
228iperf3 contains some original code from iperf2.  The authors of iperf2
229are (in alphabetical order): Jon Dugan, John Estabrook, Jim Ferbuson,
230Andrew Gallatin, Mark Gates, Kevin Gibbs, Stephen Hemminger, Nathan
231Jones, Feng Qin, Gerrit Renker, Ajay Tirumala, Alex Warshavsky.
232