This release represents eight months of development since v0.14.0. There
are several new features and a bunch of bugfixes. Also, starting from
this release, Flent will be packaged for Debian and included in upcoming
releases of both Debian and Ubuntu. Thanks to Iain Learmonth for
sponsoring this.
Changes since v0.14.0 include:
- Several new tests, including the rtt_fair_var* tests and
tcp_n{up,down} tests which take a variable number of hostnames as
targets. - Fixes to the UDP tests and a new test parameter to support setting UDP
flow bandwidth. - Support for setting TOS values for some tests.
- Added parser for WiFi statistics such as aggregation size and airtime
(the latter requires a patched ath9k). - Support globbing in test plot specifications.
- Added test mixins for adding several types of extra flows by
specifying test parameters. These include ping flows, VoIP flows, HTTP
traffic and WiFi and Qdisc stats. The parameters are documented in the
'tests' section of the man page and documentation. - Generate a UUID when running a batch sequence, and support grouping
combination plots on these. This makes it possible to turn individual
batch sequences into data series when plotting combination plots. - Support saving the intermedia result sets generated when creating
combination plots. This speeds up subsequent plottings of the same
data (which is helpful when experimenting with customising the plots). - Added a bunch of command line options to customise lots of aspects of
the plot appearance (labels, legend placement, etc.). See man page or
--help output for details. - Add a runner to parse /proc/net/netstat TCPext mib entries.
- Customise plot appearance by directly setting matplotlib RC parameters
instead of loading an RC file. This means that plots will always use
the included style unless --no-matplotlibrc is used. - Allow plotting results from different test names on the same plot.
This is especially useful with the mixins, which generate the same
series names independent on which underlying test is used. No checking
is done for whether two different tests make sense to plot together,
though, so use common sense. - Support computing Jain's fairness index as a plot combiner. This means
that it is now possible to plot the fairness index across averages of
datasets, for instance to compare average throughput values over whole
test runs. - Support computing (and plotting) MOS scores from latency and delay
values using the ITU G.107 06/2015 E-model. - Added rudimentary support for running a test runner on a remote host.
I.e. Flent can now SSH to another host and run (e.g.) a netperf
instance and add that data into a test. - Support normalising data series by each other when plotting. This can
be used to, for instance, normalise individual host throughput values
by the total throughput to get a fractional value. - Support using the raw values when reducing data series in combination
plots. - A bunch of bug fixes too numerous to list here.