CHANGES.md

Changes since latest release

Changes since v1.2.2 include:

Flent v1.2.2

Released on 2018-04-17.

Changes since v1.2.1 include:

• Add workaround for the new timeout behaviour in fping v4.0, which would result in no ping data being recorded.

• Fix plotting bugs where the idle period at the start and end of tests were not removed from aggregate plots in single-flow TCP tests.

• Add missing {bar,box}combine{up,down} plots to RRUL test.

• Fix filtering of combination plot filenames and make sure legends are filtered in box plot ticklabels.

• Add voip_marking test parameter to set diffserv marking for VoIP flows.

• Add udp_flood_var_up_staggered test that runs multiple UDP flood flows with staggered start times.

Flent v1.2.1

Released on 2018-03-28.

Changes since v1.2.0 include:

• Fix DSCP handling for irtt flows.

• Fix several crashes in the GUI.

• Fix handling of empty data series and several potential crashes in the plotting code.

• Fix potential crash in metadata gathering and add timeout to commands.

• Add several missing plots to various tests.

Flent v1.2.0

Released on 2018-02-06.

Changes since v1.1.1 include:

• Add support for the irtt binary (https://github.com/peteheist/irtt/) for isochronous UDP latency tests. If irtt is available in $PATH, it will be preferred over netperf for UDP RTT tests and over D-ITG for VoIP tests. This means that UDP latency tests will no longer use more bandwidth as the RTT decreases, and VoIP tests are easier to setup. Many thanks to Pete Heist for writing the irtt tool.

  As part of this change, a generic facility for runner preferences has been added, which makes it possible to define a test in terms of higher level functionality and let Flent pick the best available underlying tool to run the test. For now this is only used in the cases mentioned above for irtt.

• Add a configurable option for overriding the colour mode for plots. This makes it possible to change how colours are assigned to different data series.

• Improve handling of multi-value options between batch files, rc file and command line. This means that multi-value options can now use both comma and semicolon as separators in the batch file, and values can be quoted to prevent splitting.

• Drop compatibility with matplotlib versions earlier than 1.4.2. Using older versions resulted in spurious errors anyway, and it is too much backporting work to support them properly.

• Fix batch mode logging to make sure a log file for a batch run only includes log lines from that run and not previous runs.

• Fix several bugs in the plotting and runner code.

Flent v1.1.1

Released on 2017-11-15.

Changes since v1.1.0 include:

• Fix several plotting bugs resulting from the overhaul of the plotting code. This includes a couple of crash bugs, bugs in the airtime plots, and a bug where all bar plots were completely broken, showing wrong values.

Flent v1.1.0

Released on 2017-10-26.

Changes since v1.0.1 include:

• A complete overhaul of the plotting code so that it now uses the exact data points captured from the test tools wherever possible, instead of interpolating values to align data points on the time axis. This should improve the accuracy of plots, especially for integer-value data series such as packet drops.

• The GUI has gained a new widget that makes it possible to control all plot-related settings. Previously, only a few of the settings for generating plots that are available on the command line could be set from within the GUI; now, the full set of command line settings can be manipulated in the GUI settings pane as well.

• Another rewrite of the plot layout algorithm which should result in fewer issues with plot elements such as titles being drawn on top of the plot.

• More keyboard shortcuts in the GUI: Ctrl+Up/Down will move between different plots, and x/X and y/Y will zoom the axes in/out.

• By default, batch run order is now randomised to prevent periodic errors biasing results in long tests. This can be turned off with the --batch-no-shuffle option.

• Netperf is now instructed to prefill its buffers with random data, instead of the data packets consisting repeated 7-byte strings ('netperf'). This ensures that data flows are not trivially compressible, which should improve test accuracy on encapsulated connections that enable compression.

• Added support for setting TCP congestion control and diffserv markings as test parameters for some tests.

• Added support for capturing socket statistics for TCP flows and store them as an auxiliary data set. This makes it possible to capture window size and RTT estimates from the kernel TCP state machine (Linux only). Thanks to Matthias Tafelmeier for the initial implementation of this feature.

• Added support for watchdog timers to make sure runners don't go over time. This is applied to the fping runner thus far.

• More metadata is captured from test runs; in particular, Netperf TCP_INFO variables, the congestion control used for each flow and socket buffer sizes are captured for every flow and stored in the metadata object.

• Numerous bug fixes throughout.

Flent v1.0.1

Released on 2017-01-16.

Changes since v1.0.0 include:

• Change the default hosts used for rtt_fair tests. One of the old ones did not exist anymore.

• Fix a bunch of crashes and behavioural issues in the new test dialog in the GUI.

• Make sure log entries emitted during startup make it to the GUI console.

Flent v1.0.0

Released on 2017-01-14.

Changes since v0.15.0 include:

• With most of the longtime outstanding issues closed, Flent has now reached the big 1.0. Woohoo!

• Use the UltraJSON library (if available) to load data files; this speeds up loading of data files moderately.

• Improve logging of Flent's operations. The log file will now contain a proper superset of the console output, and Flent has been made slightly more verbose about what it is doing. Also, the log is now shown in a pane in the GUI.

• Fix an error breaking plots on Python 2 and some versions of matplotlib.

• Support PyQt5 in the GUI (and prefer it over PyQt4). If PyQt5 is not found, fall back to PyQt4.

• Add new SummaryFormatter that outputs mean and median values for each data series. This is the new default formatter, meaning that its output will be shown after a test run if no other formatter (or plot) is specified.

• Support multiprocessing in the GUI. When loading several plots at once, plotting will now be passed off to separate worker processes. This allows plotting to use all the available processors on the machine, and speeds up loading of many plots tremendously (initial load is sped up by an order of magnitude). This change also means that re-plotting on config changes will be done dynamically in the background, which makes the GUI more responsive.

• Make text completely black in the default colour scheme. This increases contrast, and helps legibility, especially on printed figures.

• Some internal code changes: Port command line parser from the old optparse class to the newer argparse, and fix a bunch of linter errors.

Flent v0.15.0

Released on 2016-10-01.

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.

Flent v0.14.0

Released on 2016-02-03.

This release adds mixins for capturing qdisc stats and CPU usage, adds some convenience functions to the GUI and fixes a bunch of bugs, most notably making the GUI work on Windows.

• Batch mode: Try to estimate the total runtime of a batch and print it before executing the batch itself.

• GUI: Add support for pinning a metadata item open when flipping between tabs, and support adding columns to the open files view from the metadata view. Both functions are available from the context menu when right-clicking in the metadata view.

• Add test mixins to a bunch of tests making it possible to simultaneously capture qdisc stats and/or CPU stats of one or more hosts while the test is run. This works by setting the test parameters qdisc_stats_hosts, qdisc_stats_interfaces and cpu_stats_hosts to comma-separated lists of hosts and interface names. The functionality relies on the tc_iterator and stats_iterator helper scripts.

• Add a C implementation of the tc_iterator helper script that works on OpenWrt and also gives higher polling accuracy than the shell script.

• Add a kill_timeout setting for runners in test definitions which will forcibly kill a subprocess after an interval (if it hasn't exited already).

• Add the tcp_upload_1000 test for seriously overloading things. On most systems, increasing the ulimit is necessary to run this test.

• Support globbing selectors in plot configuration when specifying datasets for a plot.

• Record number of failed runners (non-0 exit status) as a metadata key.

• GUI: Fix bugs in plot blitting on older matplotlibs, fix test running without a pre-set HOST, fix bugs in draw cache handling, and restore the GUI to a working state on Windows.

• Various bugfixes.

Flent v0.13.0

Released on 2015-11-06.

This release of Flent adds a couple of new features to the GUI, updates Iperf support and adds a bunch of bugfixes.

• GUI: For line graphs (timeseries and CDF plots), highlight data series when the mouse is hovering over the lines themselves or their corresponding legend items. This makes it easier to pick out specific data series when browsing graphs. There's a toggle to turn off this feature, since it can perform badly on slow systems and/or causes flicker in some instances on OSX.

• GUI: Experimental feature to run tests from the GUI. Fairly basic thus far, but it is possible to run a simple test from the GUI which will subsequently be loaded in the graph view.

• File format: Bump file format to v3. The file format now uses UTC timestamps everywhere internally, which is an incompatible change; hence the version bump. Old data files will be loaded assuming local times and converted appropriately on load. File name date stamps are still done in local time for the box running the test.

• Bring Iperf support up-to-date. The newest git version of Iperf (2) adds support for sub-second timestamps. Flent now detects this and can use and parse Iperf results when this support is detected. So far, no tests actually use Iperf, though.

• Flent will now capture the module version of a running Cake shaper module as part of extended metadata. TCP buffer size limit sysctls are now also captured.

• Some bug fixes related to matplotlib and Qt version compatibility and weirdness of the OSX file selector dialog.

Flent v0.12.4

Released on 2015-09-22.

This is a small bugfix release. Changes since v0.12.3:

• Packaging fixes for Debian.

• Support for Python 2.6 has officially been dropped (it had already bitrotted, so now a nicer error message is shown straight away).

• Bug in local_bind for ping commands fixed.

• More graphs for tcp_* tests and fixed labels on HTTP test graphs.

• Catch unhandled exception that caused crashed when using a null output formatter.

Flent v0.12.3

Released on 2015-08-03.

This is a minor bugfix release, fixing an important regression in the GUI preventing files from being loaded when running under Python 2.

Flent v0.12.2

Released on 2015-07-29.

This release includes a major refactoring of the plotting code, which should make it easier to extend in the future.

Other changes since v0.11.1:

• GUI restructuring: The plot settings now reside in a menu, the metadata view defaults to the bottom part of the window, and a new open files view has been added to make it easier to manage many open files at once in the GUI.

• Revert to writing data files to the current directory by default. A new parameter, -D, can optionally select a different output dir.

• The man page is now generated from the Sphinx documentation.

• Try to detect if the system ping utility produces parsable output before using it. Should give nicer error messages on OSX if fping is not installed.

• Add a new qdisc-stats test which will periodically gather qdisc statistics and a wlan-retries test which will gather WiFi retry statistics.

• New tests comparing different TCP congestion control algorithms have been added.

• Add a --replace-legend parameter to do search/replace on legends when plotting.

• A test suite has been added containing basic unittests for parts of the code. Test coverage is still fairly poor, but will be extended going forward. Run make test to run the test suite (requires the 'mock' Python library).

• VoIP tests can now show loss rates.

• A bunch of bug fixes.