Notable user-visible changes
Several bugs that could allow the iperf3 server to hang waiting for input on the control connection has been fixed. ESnet thanks Jorge Sancho Larraz from Canonical for reporting this issue. For more information, see: https://downloads.es.net/pub/iperf/esnet-secadv-2023-0002.txt.asc
A bug that caused garbled output with UDP tests on 32-bit hosts has been fixed (PR #1554, PR #1556). This bug was introduced in iperf-3.14.
A bug in counting UDP messages has been fixed (PR #1367, PR #1380).
Notable user-visible changes
A memory allocation hazard was fixed (Issue #1542/PR #1543). For more information see: https://downloads.es.net/pub/iperf/esnet-secadv-2023-0001.txt.asc
JSON output was improved, such as print JSON numbers as signed (PR #1539, Issue #1435), the exit code when doing JSON output was fixed (PR #1523), and client_api was fixed so that it still returns an error code when JSON is enabled (Issue #1405). Also, duplicate fields when using multiple streams was removed from the JSON output (#1492).
Prevent UDP packet count and operations overflow (PR #1536/Issue #1534).
Statistics are fixed when --omit is used (Issue #1489/PR #1498).
Developer-visible changes
CI builds and tests using GitHub actions have been added (PR #1519).
A fix for Android “unable to create a new stream error” was added (PR #1506).
Support for Voice Admit DSCP code point from RFC 5865 was added (PR #1490).
A fix for preventing a crash when RSA public key path doesn't exist was fixed (PR #1488/Issue #1471).
Notable user-visible changes
fq-rate (PR #1461, Issue #1366), and bidirectional flag (Issue #1428, PR #1429) were added to the JSON output.
Added support for OpenBSD including cleaning up endian handling (PR #1396) and support for TCP_INFO on systems where it was implemented (PR #1397).
Fixed bug in how TOS is set in mapped v4 (PR #1427).
Corrected documentation, such as updating binary download links and text (Issue #1459), updating version on iperf3 websites, and fixing an incorrect error message (Issue #1441).
Fixed crash on rcv-timeout with JSON logfile (#1463, #1460, issue #1360, PR #1369).
Fixed a bug that prevented TOS/DSCP from getting set correctly for reverse tests (PR #1427, Issue #638).
Developer-visible changes
Getter and setter are now available for bind_dev (PR #1419).
Added missing getter for bidirectional tests (PR #1453).
Added minor changes to clean up .gitignore and error messages (#1408).
Made sure configure scripts are runnable with /bin/sh (PR #1398).
Cleaned up RPM spec, such as adding missing RPM build dependencies, dropping EL5 and removing outdated %changelog (PR #1401) to make.
Added a fix for a resource leak bug in function iperf_create_pidfile(#1443).
Notable user-visible changes
cJSON has been updated to version 1.7.15 (#1383).
The --bind % option syntax now works properly (#1360 / #1371).
A server-side file descriptor leak with the --logfile option has been fixed (#1369 / #1360 / #1369 / #1389 / #1393).
A bug that caused some large values from TCP_INFO to be misprinted as negative numbers has been fixed (#1372).
Using the -k or -n flags with --reverse no longer leak into future tests (#1363 / #1364).
There are now various debug level options available with the --debug option. These can be used to adjust the amount of debugging output (#1327).
A new --snd-timeout option has been added to set a termination timeout for idle TCP connections (#1215 / #1282).
iperf3 is slightly more robust to out-of-order packets during UDP connection setup in --reverse mode (#914 / #1123 / #1182 / #1212 / #1260).
iperf3 will now use different ports for each direction when the --cport and --bdir options are set (#1249 / #1259).
The iperf3 server will now exit if it can't open its log file (#1225 / #1251).
Various help message and output fixes have been made (#1299 / #1330 / #1345 / #1350).
Various compiler warnings have been fixed (#1211 / #1316).
Developer-visible changes
Operation of bootstrap.sh has been fixed and simplified (#1335 / #1325).
Flow label support / compatibility under Linux has been improved (#1310).
Various minor memory leaks have been fixed (#1332 / #1333).
A getter/setter has been added for the bind_port parameter (--cport option). (#1303, #1305)
Various internal documentation improvements (#1265 / #1285 / #1304).
Notable user-visible changes
Update links to Discussions in documentation
Fix DSCP so that TOS = DSCP * 4 (#1162)
Fix --bind-dev for TCP streams (#1153)
Fix interface specification so doesn't overlap with IPv6 link-local addresses for -c and -B (#1157, #1180)
Add get/set test_unit_format function declaration to iperf_api.h
Auto adjustment of test-end condition for file transfers (-F), if no end condition is set, it will automatically adjust it to file size in bytes
Exit if idle time expires waiting for a connection in one-off mode (#1187, #1197)
Support zerocopy by reverse mode (#1204)
Update help and manpage text for #1157, support bind device
Consistently print target_bandwidth in JSON start section (#1177)
Test bitrate added to JSON output (#1168)
Remove fsync call after every write to receiving --file (#1176, #1159)
Update documentation for -w (#1175)
Fix for #952, different JSON object names for bidir reverse channel
Notable user-visible changes
Developer-visible changes
Notable user-visible changes
Fix a bug where some --reverse tests didn't terminate (#982 / #1054).
Responsiveness of control connections is slightly improved (#1045 / #1046 / #1063).
The allowable clock skew when doing authentication between client and server is now configurable with the new --time-skew-threshold (#1065 / #1070).
Bitrate throttling using the -b option now works when a burst size is specified (#1090).
A bug with calculating CPU utilization has been fixed (#1076 / #1077).
A --bind-dev option to support binding sockets to a given network interface has been added to make iperf3 work better with multi-homed machines and/or VRFs (#817 / #1089 / #1097).
--pidfile now works with --client mode (#1110).
The server is now less likely to get stuck due to network errors (#1101, #1125), controlled by the new --rcv-timeout option.
Fixed a few bugs in termination conditions for byte or block-limited tests (#1113, #1114, #1115).
Added tcp_info.snd_wnd to JSON output (#1148).
Some bugs with garbled JSON output have been fixed (#1086, #1118, #1143 / #1146).
Support for setting the IPv4 don't-fragment (DF) bit has been added with the new --dont-fragment option (#1119).
A failure with not being able to read the congestion control algorithm under WSL1 has been fixed (#1061 / #1126).
Error handling and error messages now make more sense in cases where sockets were not successfully opened (#1129 / #1132 / #1136, #1135 / #1138, #1128 / #1139).
Some buffer overflow hazards were fixed (#1134).
Notable developer-visible changes
Notable user-visible changes
A --timestamps flag has been added, which prepends a timestamp to each output line. An optional argument to this flag, which is a format specification to strftime(3), allows for custom timestamp formats (#909, #1028).
A --server-bitrate-limit flag has been added as a server-side command-line argument. It allows a server to enforce a maximum throughput rate; client connections that specify a higher bitrate or exceed this bitrate during a test will be terminated. The bitrate is expressed in bits per second, with an optional trailing slash and integer count that specifies an averaging interval over which to enforce the limit (#999).
A bug that caused increased CPU usage with the --bidir option has been fixed (#1011).
Notable developer-visible changes
Notable user-visible changes
Notable user-visible changes
Profiled libraries and binaries are no longer built by default (#950).
A minimal Dockerfile has been added (#824).
A bug with burst mode and unlimited rate has been fixed (#898).
Configuring with the --enable-static-bin flag will now cause a statically-linked iperf3 binary to be built (#989).
Configuring with the --without-sctp flag will now prevent SCTP from being auto-detected (#1008). This flag allows building a static binary (see above item) on a CentOS system with SCTP installed, because no static SCTP libraries are available.
Clock skew between the iperf3 client and server will no longer skew the computation of jitter during UDP tests (#842 / #990).
A possible buffer overflow in the authentication feature has been fixed. This was only relevant when configuration authentication using the libiperf3 API, and did not affect command-line usage. Various other improvements and fixes in this area were also made (#996).
Notable developer-visible changes
The embedded version of cJSON has been updated to 1.7.13 (#978).
Some server authentication functions have been added to the API (#911).
API access has been added to the connection timeout parameter (#1001).
Tests for some authentication functions have been added.
Various compiler errors and warnings have been fixed.
Notable user-visible changes
Support for simultaneous bidirectional tests with the --bidir flag (#201/#780).
Use POSIX standard clock_gettime(3) interface for timekeeping where available (#253/#738).
Passwords for authentication can be provided via environment variable (#815).
Specifying --repeating-payload and --reverse now works (#867).
Failed authentication doesn't count for --one-off (#864/#877).
Several memory leaks related to authenticated use were fixed (#881/#888).
The delay for tearing down the control connection for the default timed tests has been increased, to more gracefully handle high-delay paths (#751/#859).
Notable developer-visible changes
Various improvements to the libiperf APIs (#767, #775, #869, #870, #871)
Fixed build behavior when OpenSSL is absent (#854).
Portability fixes (#821/#874).
Notable user-visible changes
A new --extra-data option can be used to fill in a user-defined string field that appears in JSON output. (#600 / #729)
A new --repeating-payload option makes iperf3 use a payload pattern similar to that used by iperf2, which could help in recreating results that might be affected by payload entropy (for example, compression). (#726)
-B now works properly with SCTP tests. (#678 / #715)
A compile fix for Solaris 10 was added. (#711)
Some minor bug fixes for JSON output. In particular, warnings for debug and/or verbose modes with --json output (#737) and a fix for JSON output on CentOS 6 (#727 / #744).
software.es.net and downloads.es.net now support HTTPS, so URLs in documentation that refer to those two hosts now use https:// instead of http:// URLs. (#759)
Notable developer-visible changes
Functions related to authenticated iperf3 connections have been exposed via libiperf. (#712 / #713)
The ToS byte is now exposed in the libiperf API. (#719)
Notable user-visible changes
Notable user-visible changes
The -A (set processor affinity) command-line flag is now supported on Windows (#665).
iperf3 now builds on systems lacking a daemon(3) library call (#369).
A bug in time skew checking under authentication was fixed (#674).
IPv6 flow labels now work correctly with multiple parallel streams (#694).
The client no longer closes its control connection before sending end-of-test statistics to the server (#677). This fixes a regression introduced in iperf-3.2.
Sending output to stdout now makes errors go to stderr, as per UNIX convention (#695).
A server side crash in verbose output with a client running multiple parallel connections has been fixed (#686).
The --cport option can now be specified without the --bind option. Using the --cport option on Linux can eliminate a problem with ephemeral port number allocation that can make multi-stream iperf3 tests perform very poorly on LAGG links. Also, the --cport option now works on SCTP tests (#697).
Notable developer-visible changes
iperf3 now builds on (some) macOS systems older than 10.7 (#607).
Some unused code and header inclusions were eliminated (#667, #668). Also some code was cleaned up to eliminate (or at least reduce) compiler warnings (#664, #671).
Notable user-visible changes
iperf3 can now be built --without-openssl on systems where OpenSSL is present (#624, #633).
A bug with printing very large numbers has been fixed (#642).
A bug where the server would, under certain circumstances, halt a test after exactly fifteen seconds has been fixed (#645).
The --tos parameter is no longer “sticky” between tests when doing --reverse tests (#639).
The authentication token on the server is properly reset between tests (#650).
A bug that could cause iperf3 to overwrite the PID file of an already-existing iperf3 process has been fixed (#623).
iperf3 will now ignore nonsensical TCP MSS values (from the TCP control connection) when trying to determine a reasonable block size for UDP tests. This condition primarily affected users on Windows, but potentially improves robustness for all platforms. (#659)
Notable developer-visible changes
User-visible changes
Authentication via a username/password mechanism, coupled with a public-key pair, is now an optional way of limiting access to an iperf3 server (#517).
Ending statistics are less ambiguous for UDP and also now use correct test durations for all protocols (#560, #238). Many fixes have been made in statistics printing code, generally for human-readable output (#562, #575, #252, #443, #236).
Several problems with the -F/--file options have been fixed. Documentation has been improved to note some ways in which this feature might not behave as expected (#588).
iperf3 now uses the correct “bitrate” phraseology rather than “bandwidth” when describing measurement results. The --bandwidth option has been renamed --bitrate, although --bandwidth is still accepted for backwards compatibility (#583).
Application-level bandwidth pacing (--bitrate option) is now checked every millisecond by default, instead of of every tenth of a second, to provide smoother traffic behavior when using application pacing (#460). The pacing can be tuned via the use of the --pacing-timer option (#563).
A new --dscp option allows specifying the DSCP value to be used for outgoing packets (#508). The TOS byte value is now printed in the JSON output (#226).
Congestion window data on FreeBSD is now computed correctly (#465, #475, #338).
The T/t suffixes for terabytes/terabits are now accepted for quantities where suffixes are supported, such as --bandwidth (#402).
Sanity checks for UDP send sizes have been added (#390), and existing checks on the --window option have been improved (#557).
The TCP rttvar value is now available in the JSON output (#534), as are the socket buffer sizes (#558).
Error handling and documentation have been improved for the -f/--format options (#568).
A new --connect-timeout option on the client allows specifying a length of time that the client will attempt to connect to the server, in milliseconds (#216).
The hostname and current timestamp are no longer used in the cookie used to associate the client and server. Instead, random data is used. Note that iperf3 now requires the /dev/urandom device (#582).
Prior versions of iperf3 doing UDP tests used to overcount packet losses in the presence of packet reordering. This has been (partially) fixed by try to not count the sequence number gaps resulting from out-of-order packets as actual losses (#457).
iperf3 no longer prints results from very small intervals (10% of the statistics reporting interval) at the end of the test run if they contain no data. This can happen due to timing difference or network queueing on the path between the client and server. This is primarily a cosmetic change to prevent these fairly meaningless intervals from showing up in the output (#278).
Compatibility note: Users running iperf3 3.2 or newer from the bwctl utility will need to obtain version 1.6.3 or newer of bwctl. Note that bwctl, a component of the perfSONAR toolkit, has been deprecated in favor of pScheduler since the release of perfSONAR 4.0.
Developer-visible changes
Various warnings and build fixes (#551, #564, #518, #597).
Some improvements have been made for increased compatibility on IRIX (#368) and with C++ (#587).
cJSON has been updated to 1.5.2 (#573), bringing in a number of bugfixes.
Some dead code has been removed.
iperf 3.1.7 is functionally identical to iperf 3.1.6. Its only changes consist of updated documentation files and text in the RPM spec file.
The release notes for iperf 3.1.6 describe changes, including bug fixes and new functionality, made since iperf 3.1.5.
User-visible changes
Specifying --fq-rate or --no-fq-socket-pacing on a system where these options are not supported now generate an error instead of a warning. This change makes diagnosing issues related to pacing more apparent.
Fixed a bug where two recently-added diagnostic messages spammed the JSON output on UDP tests.
The release notes for iperf 3.1.5 describe changes, including bug fixes and new functionality, made since iperf 3.1.4.
Compatibility note: Fair-queueing is now specified differently in iperf 3.1.5 than in prior versions (which include 3.1.3 and 3.1.4).
Compatibility note: UDP tests may yield different results from all prior versions of iperf3 (through 3.1.4) due to the new default UDP sending size.
User-visible changes
The fair-queueing per-socket based pacing available on recent Linux systems has been reimplemented with a different user interface (#325, #467, #488). The --bandwidth command-line flag now controls only the application-level pacing, as was the case in iperf 3.1.2 and all earlier versions. Fair-queueing per-socket based pacing is now controlled via a new --fq-rate command-line flag. Note that TCP and UDP tests may use --bandwidth, --fq-rate, both flags, or neither flag. SCTP tests currently support --bandwidth only. The --no-fq-socket-pacing flag, which was introduced in iperf 3.1.3, has now been deprecated, and is equivalent to --fq-rate=0. iperf3 now reacts more gracefully if --no-fq-socket-pacing or --fq-rate are specified on platforms that don't support these options.
For UDP tests, note that the default --bandwidth is 1 Mbps. Using the fair-queueing-based pacing will probably require explicitly setting both --bandwidth and --fq-rate, preferably to the same value. (While setting different values for --bandwidth and --fq-rate can certainly be done, the results can range from entertaining to perplexing.)
iperf3 now chooses a more sane default UDP send size (#496, #498). The former default (8KB) caused IP packet fragmentation on paths having smaller MTUs (including any Ethernet network not configured for jumbo frames). This could have effects ranging from increased burstiness, to packet loss, to complete failure of the test. iperf3 now attempts to use the MSS of the control connection to determine a default UDP send size if no sending length was explicitly specified with --length.
Several checks are now made when setting the socket buffer sizes with the -w option, to verify that the settings have been applied correctly. The results of these checks can been seen when the --debug flag is specified. (#356)
A --forceflush flag has been added to flush the output stream after every statistics reporting interval.
Developer-visible changes
The release notes for iperf 3.1.4 describe changes, including bug fixes and new functionality, made since iperf 3.1.3.
User-visible changes
Developer-visible changes
Fixed a buffer overflow in the cJSON library (issue #466). It is not believed that this bug created any security vulnerabilities in the context of iperf3.
Travis CI builds are now enabled on this codeline (pull request #424).
Various bug fixes (issue #459, pull request #429, issue #388).
The release notes for iperf 3.1.3 describe changes, including bug fixes and new functionality, made since iperf 3.1.2.
Security
User-visible changes
On supported platforms (recent Linux), iperf3 can use fair-queueing-based per-socket pacing instead of its own application-level pacing for the --bandwidth option. Application-level pacing can be forced with the -no-fq-socket-pacing flag.
A bug that could show negative loss counters with --udp and --omit has been fixed (issue #412, pull request #414).
Error handling has been made slightly more robust. Also, the iperf3 server will no longer exit after five consecutive errors, but will only exit for certain types of errors that prevent it from participating in any tests at all.
Developer-visible changes
Fixed the build on FreeBSD 11-CURRENT (issue #413).
Fixed various coding errors (issue #423, issue #425).
The release notes for iperf 3.1.2 describe changes, including bug fixes and new functionality, made since iperf 3.1.1.
User-visible changes
Fixed a bug that caused nan values to be emitted (incorrectly) into JSON, particularly at the end of UDP tests (issue #278).
Fixed a bug that caused the wrong value to be printed for out-of-order UDP packets (issue #329).
Added a contrib/ directory containing a few submitted graphing scripts.
Developer-visible changes
The release notes for iperf 3.1.1 describe changes and new functionality in iperf 3.1.1, but not present in 3.1.
User-visible changes
Some markup fixes have been made in the manpages for Debian compatibility (issue #291).
A bug where the -T title option was not being output correctly in JSON output has been fixed (issue #292).
Argument handling for some command-line options has been improved (issue #316).
Developer-visible changes
The release notes for iperf 3.1 describe changes and new functionality in iperf 3.1, but not present in 3.0.11 or any earlier 3.0.x release.
Selected user-visible changes
SCTP support has been added (with the --sctp flag), on Linux, FreeBSD, and Solaris (issue #131).
Setting CPU affinity now works on FreeBSD.
Selection of TCP congestion now works on FreeBSD, and is now called --congestion (the old --linux-congestion option works but is now deprecated).
A new -I option for the server causes it to write a PID file, mostly useful for daemon mode (issue #120).
A --logfile argument can now force all output to go to a file, rather than to a file. This is especially useful when running an iperf3 server in daemon mode (issue #119).
Various compatibility fixes for Android (issue #184, issue #185), iOS (issue #288), NetBSD (issue #248), Solaris (issue #175, issue #178, issue #180, issue #211), vxWorks (issue #268).
A --udp-counters-64bit flag has been added to support very long-running UDP tests, which could cause a counter to overflow (issue #191).
A --cport option to specify the client-side port has been added (issue #207, issue #209, issue #239).
Some calculation errors with the -O feature have been fixed (issue #236).
A potential crash in the iperf3 server has been fixed (issue #257, issue #258).
Many miscellaneous bug fixes.
Selected developer-visible changes
Consumers of libiperf can now get the JSON output for a just-completed test (issue #147).
Detection of various optional features has been improved to check for the presence or absence of platform functionality, not the name of platforms.
Out-of-tree builds now work (issue #265).
User-visible changes
Added -1 / --one-off flag, which causes the iperf3 server to process one client connection and then exit. Intended primarily for bwctl integration (issue #230).
Added various minor bug fixes (issues #231, #232, #233).
Added 30-second timeout for UDP tests if unable to establish UDP connectivity between sender and receiver (issue #222).
User-visible changes
Fixed the build on MacOS X Yosemite (issue #213).
UDP tests now honor the -w option for setting the socket buffer sizes (issue #219).
Developer-visible changes
Added an RPM spec file plus functionality to fill in the version number.
Fixed potential filename collision with a system header (issue #203).
User-visible changes
Developer-visible changes
User-visible changes
Developer-visible changes
User-visible changes
A server bug where new connections from clients could disrupt running tests has been fixed (issue #202).
Rates now consistently use 1000-based prefixes (K, M, G), where sizes of objects now consistently use 1024-based prefixes (issue #173).
UDP tests with unlimited bandwidth are now supported (issue #170).
An interaction between the -w and -B options, which kept them from working when used together, has been fixed (issue #193).
Developer-visible changes
User-visible changes
Several bugs that kept the -B option from working in various circumstances have been fixed (issue #193).
Various compatibility fixes for OpenBSD (issue #196) and Solaris (issue #177).
Developer-visible changes
User-visible changes
Erroneous output when doing --json output has been fixed (this problem was caused by an attempt to fix issue #158).
The maximum test running time has been increased from one hour to one day (issue #166).
Project documentation has been moved to GitHub Pages at this URL: http://software.es.net/iperf/.
A bug that caused CPU time to be computed incorrectly on FreeBSD has been fixed.
A timing issue which caused measurement intervals to be wrong with TCP tests on lossy networks has been fixed (issue #125).
Newer versions of autoconf / automake / libtool are now used by default (issue #161).
JSON output now indicates whether the test was run in --reverse mode (issue #167).
It is now possible to get (most of) the server-side output at the client by using the --get-server-output flag (issue #160).
Developer-visible changes
User-visible changes
Due to several oversights, the source code archive for iperf 3.0.2 was distributed as an uncompressed tarball, despite having an extension (“.tar.gz”) that indicated it was compressed. The release generation procedure has been changed to avoid this problem going forward.
Summary structures in the JSON output are now included, even if there is only one stream. This change makes consuming the JSON output easier and more consistent (issue #151).
A possible buffer overflow in iperf_error.c has been fixed (issue #155).
Developer-visible changes
User-visible changes
The iperf3 project has been moved to GitHub, and various URLs in documentation files have been changed to point there.
iperf3 now builds on Linux systems that do not support TCP_CONGESTION. Most notably this allows iperf3 to work on CentOS 5.
An abort on MacOS 10.9 has been fixed (issue #135).
Added -I flag for the server to write a PID file, mostly useful for daemon mode (issue #120).
A bug that could break some TCP tests on FreeBSD has been fixed.
TCP snd_cwnd output is now printed by default on Linux (issue #99).
In JSON output, the --title string no longer has a colon and two spaces appended (issue #139).
A buffer for holding formatted numeric values is now properly-sized so that output is not truncated (issue #142).
Developer-visible changes
Some memory leaks have been fixed.
A -d flag enables debugging output.
A .gitignore file has been added.
libtoolize is now invoked correctly from the bootstrap.sh script.
The test unit format can now be set from the API (issue #144).
libiperf is now built as both shared and static libraries.
In the JSON output, the “connection” structures are now stored as an array in the “start” block, instead of overwriting each other. While technically an incompatible API change, the former behavior generated unusable JSON.