aboutsummaryrefslogtreecommitdiffstats
path: root/resources/tools/presentation/doc/graphs_improvements.rst
diff options
context:
space:
mode:
authorpmikus <peter.mikus@protonmail.ch>2023-05-24 13:24:19 +0000
committerpmikus <peter.mikus@protonmail.ch>2023-05-24 13:24:19 +0000
commitfeac1add7b15bb7d66da1320bb6a6e95a722c504 (patch)
tree445bf87a249b7a63916a40055dd9b1737666a782 /resources/tools/presentation/doc/graphs_improvements.rst
parentd164bef0373edfd2b6cc7d4aaa27b928065df3e5 (diff)
remove(tools): presentation, docs
Signed-off-by: pmikus <peter.mikus@protonmail.ch> Change-Id: Icc07d18b9c0bd00af157bb817205e5b54824d525
Diffstat (limited to 'resources/tools/presentation/doc/graphs_improvements.rst')
-rw-r--r--resources/tools/presentation/doc/graphs_improvements.rst590
1 files changed, 0 insertions, 590 deletions
diff --git a/resources/tools/presentation/doc/graphs_improvements.rst b/resources/tools/presentation/doc/graphs_improvements.rst
deleted file mode 100644
index 331905cabb..0000000000
--- a/resources/tools/presentation/doc/graphs_improvements.rst
+++ /dev/null
@@ -1,590 +0,0 @@
-================================
- Envisioning information by PAL
-================================
-
-Introduction
-------------
-
-This document describes possible improvements in data presentation provided by
-PAL for the `Report <https://s3-docs.fd.io/csit/master/report/>`_ and the
-`Trending <https://s3-docs.fd.io/csit/master/trending/>`_
-
-You can generate a standalone html version of this document using e.g.
-rst2html5 tool:
-
-.. code:: bash
-
- rst2html5 --stylesheet graphs_improvements.css graphs_improvements.rst >> graphs_improvements.html
-
-**Modifications of existing graphs**
-
-- `Speedup Multi-core`_
-- `Packet Throughput`_
-- `Packet Latency`_
-
-**New graphs to be added**
-
-- `Comparison between releases`_
-- `Comparison between processor architectures`_
-- `Comparison between 2-node and 3-node topologies`_
-- `Comparison between different physical testbed instances`_
-- `Comparison between NICs`_
-- `Other comparisons`_
-
-**Export of static images**
-
-- low priority
-- make possible to `export static images`_ which are available via link on the
- web page.
-- vector formats (svg, pdf) are preferred
-
-Priorities
-----------
-
-**Target CSIT-18.10**
-
-- `Speedup Multi-core`_
-- `Packet Throughput`_
-
-**Nice to have in CSIT-18.10**
-
-.. note::
-
- Only if above done, and target CSIT-18.10 is in time , otherwise next
- release.
-
-- `Packet Latency`_
-
-Modifications of existing graphs
---------------------------------
-
-The proposed modifications include the changes in:
-
-- the layout of the graphs,
-- the data and way how it is presented,
-- the test cases presented in the graphs.
-
-The first two points are described below, the last one will be added later as a
-separate chapter.
-
-.. _Speedup Multi-core:
-
-Speedup Multi-core
-``````````````````
-
-The "Speedup Multicore" graph will display the measured data together with
-perfect values calculated as multiples of the best value measured using one
-core. The relative difference between measured and perfect values will be
-displayed in the hover next to each data point.
-
-.. image:: pic/graph-speedup.svg
- :width: 800 px
- :scale: 50 %
- :align: center
- :alt: Graph "Speedup Multi-core: not found.
-
-**Description:**
-
-*Data displayed:*
-
-- one or more data series from the same area, keep the number of displayed
- data series as low as possible (max 6)
-- x-axis: number of cores
-- y-axis: throughput (measured and perfect) [Mpps], linear scale, beginning
- with 0
-- hover information: Throughput [Mpps], Speedup [1], Relative difference between
- measured and ideal values [%], Perfect Throughput [%]
-- Limits of ethernet links, NICs and PCIe. See `Physical performance limits`_.
-
-*Layout:*
-
-- plot type: lines with data points (plotly.graph_objs.Scatter)
-- data series format:
- - measured: solid line with data points
- - perfect: dashed line with data points, the same color as "measured"
-- title: "Speedup Multi-core: <area, scaling, features, ...>",
- top, centered, font size 18; configurable in specification file: visible /
- hidden, text
-- x-axis: integers, starting with 1 (core), linear, font size 16, bottom
-- x-axis label: "Number of cores [qty]", bottom, centered, font size 16
-- y-axis: float, starting with 0, dynamic range, linear, font size 16, left
-- y-axis label: "Packet Throughput [Mpps]", middle, left
-- legend: list of presented test cases, bottom, left, font size 16; the order
- of displayed tests is configurable in the specification file
-- annotation: text: "dashed: perfect<br>solid: measured", top, left,
- font size 16
-
-.. _Packet Throughput:
-
-Packet Throughput
-`````````````````
-
-The "Packet Throughput" graph will display the measured data using
-statistical box graph. Each data point is constructed from 10 samples.
-The statistical data are displayed as hover information.
-
-.. image:: pic/graph-throughput.svg
- :width: 800 px
- :scale: 50 %
- :align: center
- :alt: Graph "Packet Throughput" not found.
-
-**Description:**
-
-*Data displayed:*
-
-- one or more data points from the same area, keep the number of displayed
- data points as low as possible (max 6)
-- x-axis: indexed test cases
-- y-axis: throughput [Mpps], linear scale, beginning with 0
-- hover information: statistical data (min, lower fence, q1, median, q3,
- higher fence, max), test case name
-
-*Layout:*
-
-- plot type: statistical box (plotly.graph_objs.Box)
-- data series format: box
-- title: "Packet Throughput: <area, scaling, features, framesize, cores, ...>",
- top, centered, font size 18; configurable in specification file: visible /
- hidden, text
-- x-axis: integers, starting with 1, linear, font size 16, bottom; the order
- of displayed tests is configurable in the specification file
-- x-axis label: "Indices of Test Cases [Index]", bottom, centered, font size 16
-- y-axis: floats, starting with 0, dynamic range, linear, font size 16, left
-- y-axis label: "Packet Throughput [Mpps]", middle, left
-- legend: "Indexed Test Cases [idx]", bottom, left, font size 16
-
-.. _Packet Latency:
-
-Packet Latency
-``````````````
-
-The "Packet Latency" graph will display the measured data using
-statistical box graph. Each data point is constructed from 10 samples.
-The statistical data are displayed as hover information.
-
-.. image:: pic/graph-latency.svg
- :width: 800 px
- :scale: 50 %
- :align: center
- :alt: Graph "Packet Latency" not found.
-
-**Description:**
-
-*Data displayed:*
-
-- one or more data points from the same area, keep the number of displayed
- data points as low as possible (max 6)
-- x-axis: data flow directions
-- y-axis: latency min/avg/max [uSec], linear scale, beginning with 0
-- hover information: statistical data (min, avg, max), test case name, direction
-
-*Layout:*
-
-- plot type: scatter with errors (plotly.graph_objs.Scatter)
-- data series format: data point with min amd max values
-- title: "Packet Latency: <area, scaling, features, framesize, cores, ...>",
- top, centered, font size 18; configurable in specification file: visible /
- hidden, text
-- x-axis: text, font size 16, bottom; the order of displayed tests is
- configurable in the specification file
-- x-axis label: "Direction", bottom, centered
-- y-axis: integers, starting with 0, dynamic range, linear, font size 16, left
-- y-axis label: "Packet Latency min/avg/max [uSec]", middle, left
-- legend: "Indexed Test Cases [idx]", bottom, left, font size 16
-
-New graphs to be added
-----------------------
-
-- *Comparison between releases*
-
- - compare MRR, NDR, PDR between releases
- - use as many releases as available
-
-- *Comparison between processor architectures*
-
- - compare MRR, NDR, PDR between processor architectures
- - HSW vs SKX (vs ARM when available)
-
-- *Comparison between 2-node and 3-node topologies*
-
- - compare MRR, NDR, PDR between topologies
- - 3n-skx vs 2n-skx
-
-- *Comparison between different physical testbed instances*
-
- - compare the results of the same test (MRR, NDR, PDR) run on different
- instances of the same testbed, e.g. HSW
- - HSW vs HSW, SKX vs SKX
-
-- *Comparison between NICs*
-
- - compare the results of the same test (MRR, NDR, PDR) run on different NICs
- but on the same instance of a physical testbed.
- - x520 vs x710 vs xl710 on HSW
- - x710 vs xxv710 on SKX
-
-- *Other comparisons*
-
-.. note::
-
- - Partially based on the existing tables in the Report
- - Only selected TCs
-
-.. _Comparison between releases:
-
-Comparison between releases
-```````````````````````````
-
-This graph will compare the results of the same test from different releases.
-One graph can present the data from one or more tests logically grouped. See
-`Grouping of tests in graphs`_ for more information.
-Each data point is constructed from 10 samples. The statistical data are
-displayed as hover information.
-
-.. image:: pic/graph_cmp_releases.svg
- :width: 800 px
- :scale: 50 %
- :align: center
- :alt: Graph "Comparison between releases" not found.
-
-**Description:**
-
-*Data displayed:*
-
-- data: packet throughput
-- x-axis: release
-- y-axis: packet throughput [Mpps], linear scale, beginning with 0
-- hover information: statistical data (median, stdev), test case name, release
-
-*Layout:*
-
-- plot type: scatter with line
-- data series format: line with markers
-- title: "Packet Throughput: <area, scaling, features, framesize, cores, ...>",
- top, centered, font size 18
-- x-axis: strings, font size 16, bottom
-- x-axis label: "Release", bototm, centered, font size 16
-- y-axis: floats, starting with 0, dynamic range, linear, bottom, font size 16
-- y-axis label: "Packet Throughput [Mpps]", middle, left, font size 16
-- legend: "Test Cases", bottom, left, font size 16
-
-.. _Comparison between processor architectures:
-
-Comparison between processor architectures
-``````````````````````````````````````````
-
-This graph will compare the results of the same test from the same release run
-on the different processor architectures (HSW, SKX, later ARM).
-One graph can present the data from one or more tests logically grouped. See
-`Grouping of tests in graphs`_ for more information.
-Each data point is constructed from 10 samples. The statistical data are
-displayed as hover information.
-
-.. image:: pic/graph_cmp_arch.svg
- :width: 800 px
- :scale: 50 %
- :align: center
- :alt: Graph "Comparison between processor architectures" not found.
-
-**Description:**
-
-*Data displayed:*
-
-- data: packet throughput
-- x-axis: processor architecture
-- y-axis: throughput [Mpps], linear scale, beginning with 0
-- hover information: statistical data (median, stdev), test case name, processor
- architecture
-
-*Layout:*
-
-- plot type: scatter with line
-- data series format: line with markers
-- title: "Packet Throughput: <area, scaling, features, framesize, cores, ...>",
- top, centered, font size 18
-- x-axis: strings, font size 16, bottom, centered
-- x-axis label: "Processor architecture", bottom, centered, font size 16
-- y-axis: floats, starting with 0, dynamic range, linear, font size 16, left
-- y-axis label: "Packet Throughput [Mpps]", middle, left
-- legend: "Test cases", bottom, left, font size 16
-
-.. _Comparison between 2-node and 3-node topologies:
-
-Comparison between 2-node and 3-node topologies
-```````````````````````````````````````````````
-
-This graph will compare the results of the same test from the same release run
-on the same processor architecture but different topologies (3n-skx, 2n-skx).
-One graph can present the data from one or more tests logically grouped. See
-`Grouping of tests in graphs`_ for more information.
-Each data point is constructed from 10 samples. The statistical data are
-displayed as hover information.
-
-.. image:: pic/graph_cmp_topo.svg
- :width: 800 px
- :scale: 50 %
- :align: center
- :alt: Graph "Comparison between 2-node and 3-node topologies" not found.
-
-**Description:**
-
-*Data displayed:*
-
-- data: packet throughput
-- x-axis: topology
-- y-axis: throughput [Mpps], linear scale, beginning with 0
-- hover information: statistical data (median, stdev), test case name, topology
-
-*Layout:*
-
-- plot type: scatter with line
-- data series format: line with markers
-- title: "Packet Throughput: <area, scaling, features, framesize, cores, ...>",
- top, centered, font size 18
-- x-axis: strings, font size 16, bottom, centered
-- x-axis label: "Topology", bottom, centered, font size 16
-- y-axis: floats, starting with 0, dynamic range, linear, font size 16, left
-- y-axis label: "Packet Throughput [Mpps]", middle, left, font size 16
-- legend: "Test cases", bottom, left, font size 16
-
-.. _Comparison between different physical testbed instances:
-
-Comparison between different physical testbed instances
-```````````````````````````````````````````````````````
-
-This graph will compare the results of the same test from the same release run
-on the same processor architecture, the same topology but different physical
-testbed instances.
-One graph can present the data from one or more tests logically grouped. See
-`Grouping of tests in graphs`_ for more information.
-Each data point is constructed from 10 samples. The statistical data are
-displayed as hover information.
-
-
-.. image:: pic/graph_cmp_testbed.svg
- :width: 800 px
- :scale: 50 %
- :align: center
- :alt: Graph "Comparison between different physical testbed instances" not
- found.
-
-**Description:**
-
-*Data displayed:*
-
-- data: packet throughput
-- x-axis: physical testbed instances
-- y-axis: throughput [Mpps], linear scale, beginning with 0
-- hover information: statistical data (median, stdev), test case name, physical
- testbed instance
-
-*Layout:*
-
-- plot type: scatter with line
-- data series format: line with markers
-- title: "Packet Throughput: <area, scaling, features, framesize, cores, ...>",
- top, centered, font size 18
-- x-axis: strings, font size 16, bottom, centered
-- x-axis label: "Physical Testbed Instance", bottom, centered, font size 16
-- y-axis: floats, starting with 0, dynamic range, linear, font size 16, left
-- y-axis label: "Packet Throughput [Mpps]", middle, left, font size 16
-- legend: "Test cases", bottom, left, font size 16
-
-.. _Comparison between NICs:
-
-Comparison between NICs
-```````````````````````
-
-This graph will compare the results of the same test from the same release run
-on the same processor architecture, the same topology but different NICs.
-One graph can present the data from one or more tests logically grouped. See
-`Grouping of tests in graphs`_ for more information.
-Each data point is constructed from 10 samples. The statistical data are
-displayed as hover information.
-
-.. image:: pic/graph_cmp_nics.svg
- :width: 800 px
- :scale: 50 %
- :align: center
- :alt: Graph "Comparison between NICs" not found.
-
-**Description:**
-
-*Data displayed:*
-
-- data: packet throughput
-- x-axis: NICs
-- y-axis: packet throughput [Mpps], linear scale, beginning with 0
-- hover information: statistical data (median, stdev), test case name, NIC
-
-*Layout:*
-
-- plot type: scatter with line
-- data series format: line with markers
-- title: "Packet Throughput: <area, scaling, features, framesize, cores, ...>",
- top, centered, font size 18
-- x-axis: strings, font size 16, bottom
-- x-axis label: "NIC", bottom, centered, font size 16
-- y-axis: floats, starting with 0, dynamic range, linear, font size 16, left
-- y-axis label: "Packet Throughput [Mpps]", middle, left, font size 16
-- legend: "Test cases", bottom, left, font size 16
-
-.. _Other comparisons:
-
-Other comparisons
-`````````````````
-
-**Other tests results comparisons**
-
-- compare packet throughput for vhost vs memif
-
-**Other views on collected data**
-
-- per `Vratko Polak email on csit-dev <https://lists.fd.io/g/csit-dev/message/3008>`_.
-
-.. _Grouping of tests in graphs:
-
-Grouping of tests in graphs
----------------------------
-
-A graph can present results of one or more tests. The tests are grouped
-according to the defined criteria. In the ideal case, all graphs use the same
-groups of tests.
-
-The grouping of tests is described in a separate document.
-
-.. TODO: [MK], [TF]: Create the document.
-.. TODO: [TF]: Add the link.
-.. TODO: [TF]: Remove/edit the next paragraph when the document is ready.
-
-**Example of data grouping:**
-
-- ip4: ip4base, ip4scale20k, ip4scale200k, ip4scale2m
- - data presented in this order from left to right
-- ip6: similar to ip4
-- l2bd: similar to ip4.
-
-.. _Sorting of tests presented in a graph:
-
-Sorting of tests presented in a graph
--------------------------------------
-
-It is possible to specify the order of tests (suites) on the x-axis presented in
-a graph:
-
-- `Packet Throughput`_
-- `Packet Latency`_
-
-It is possible to specify the order of tests (suites) in the legend presented in
-a graph:
-
-- `Speedup Multi-core`_
-
-In both cases the order is defined in the specification file for each plot
-separately, e.g.:
-
-.. code:: yaml
-
- -
- type: "plot"
- <other parameters>
- sort:
- - "IP4BASE"
- - "FIB_20K"
- - "FIB_200K"
- - "FIB_2M"
-
-The sorting is based on tags. If more then one test has the same tag, only the
-first one is taken. The remaining tests and the tests without listed tags are
-placed at the end of the list in random order.
-
-.. _export static images:
-
-Export of static images
------------------------
-
-Not implemented yet. For more information see:
-
-- `Plotly: Static image export <https://plot.ly/python/static-image-export/>`_
-- prefered vector formats (svg, pdf)
-- requirements:
- - plotly-orca
- - `Orca <https://github.com/plotly/orca>`_
- - `Orca releases <https://github.com/plotly/orca/releases>`_
- - `Orca management in Python <https://plot.ly/python/orca-management/>`_
- - psutil
-
-.. _Physical performance limits:
-
-Physical performance limits
----------------------------
-
-+-----------------+----------------+
-| Ethernet links | pps @64B |
-+=================+================+
-| 10ge | 14,880,952.38 |
-+-----------------+----------------+
-| 25ge | 37,202,380.95 |
-+-----------------+----------------+
-| 40ge | 59,523,809.52 |
-+-----------------+----------------+
-| 100ge | 148,809,523.81 |
-+-----------------+----------------+
-
-
-+-----------------+----------------+
-| Ethernet links | bps |
-+=================+================+
-| 64B | |
-+-----------------+----------------+
-| IMIX | |
-+-----------------+----------------+
-| 1518B | |
-+-----------------+----------------+
-| 9000B | |
-+-----------------+----------------+
-
-
-+-----------------+----------------+
-| NIC | pps @64B |
-+=================+================+
-| x520 | 24,460,000 |
-+-----------------+----------------+
-| x710 | 35,800,000 |
-+-----------------+----------------+
-| xxv710 | 35,800,000 |
-+-----------------+----------------+
-| xl710 | 35,800,000 |
-+-----------------+----------------+
-
-
-+-----------------+----------------+
-| NIC | bw ??B |
-+=================+================+
-| x520 | |
-+-----------------+----------------+
-| x710 | |
-+-----------------+----------------+
-| xxv710 | |
-+-----------------+----------------+
-| xl710 | |
-+-----------------+----------------+
-
-
-+-----------------+----------------+
-| PCIe | bps |
-+=================+================+
-| PCIe Gen3 x8 | 50,000,000,000 |
-+-----------------+----------------+
-| PCIe Gen3 x16 | 100,000,000,000|
-+-----------------+----------------+
-
-
-+-----------------+----------------+
-| PCIe | pps @64B |
-+=================+================+
-| PCIe Gen3 x8 | 74,404,761.90 |
-+-----------------+----------------+
-| PCIe Gen3 x16 | 148,809,523.81 |
-+-----------------+----------------+