[v7,4/4] doc: add howto guide for lcore poll busyness

Message ID 20220914092929.1159773-5-kevin.laatz@intel.com (mailing list archive)
State Rejected, archived
Delegated to: David Marchand
Headers
Series Add lcore poll busyness telemetry |

Checks

Context Check Description
ci/checkpatch success coding style OK
ci/Intel-compilation fail Compilation issues
ci/iol-mellanox-Performance success Performance Testing PASS
ci/iol-intel-Performance success Performance Testing PASS
ci/iol-aarch64-unit-testing success Testing PASS
ci/iol-intel-Functional success Functional Testing PASS
ci/iol-x86_64-compile-testing success Testing PASS
ci/iol-aarch64-compile-testing success Testing PASS
ci/github-robot: build success github build: passed
ci/iol-x86_64-unit-testing success Testing PASS
ci/intel-Testing success Testing PASS

Commit Message

Kevin Laatz Sept. 14, 2022, 9:29 a.m. UTC
  Add a new section to the howto guides for using the new lcore poll
busyness telemetry endpoints and describe general usage.

Signed-off-by: Kevin Laatz <kevin.laatz@intel.com>

---
v6:
  * Add mention of perf autotest in note mentioning perf impact.

v4:
  * Include note on perf impact when the feature is enabled
  * Add doc to toctree
  * Updates to incorporate changes made earlier in the patchset

v3:
  * Update naming to poll busyness
---
 doc/guides/howto/index.rst               |  1 +
 doc/guides/howto/lcore_poll_busyness.rst | 93 ++++++++++++++++++++++++
 2 files changed, 94 insertions(+)
 create mode 100644 doc/guides/howto/lcore_poll_busyness.rst
  

Patch

diff --git a/doc/guides/howto/index.rst b/doc/guides/howto/index.rst
index bf6337d021..0a9060c1d3 100644
--- a/doc/guides/howto/index.rst
+++ b/doc/guides/howto/index.rst
@@ -21,3 +21,4 @@  HowTo Guides
     debug_troubleshoot
     openwrt
     avx512
+    lcore_poll_busyness
diff --git a/doc/guides/howto/lcore_poll_busyness.rst b/doc/guides/howto/lcore_poll_busyness.rst
new file mode 100644
index 0000000000..be5ea2a85d
--- /dev/null
+++ b/doc/guides/howto/lcore_poll_busyness.rst
@@ -0,0 +1,93 @@ 
+..  SPDX-License-Identifier: BSD-3-Clause
+    Copyright(c) 2022 Intel Corporation.
+
+Lcore Poll Busyness Telemetry
+=============================
+
+The lcore poll busyness telemetry provides a built-in, generic method of gathering
+lcore utilization metrics for running applications. These metrics are exposed
+via a new telemetry endpoint.
+
+Since most DPDK APIs polling based, the poll busyness is calculated based on
+APIs receiving 'work' (packets, completions, events, etc). Empty polls are
+considered as idle, while non-empty polls are considered busy. Using the amount
+of cycles spent processing empty polls, the busyness can be calculated and recorded.
+
+Application Specified Busyness
+------------------------------
+
+Improved accuracy of the reported busyness may need more contextual awareness
+from the application. For example, an application may make a number of calls to
+rx_burst before processing packets. If the last burst was an "empty poll", then
+the processing time of the packets would be falsely considered as "idle", since
+the last burst was empty. The application should track if any of the polls
+contained "work" to do and should mark the 'bulk' as "busy" cycles before
+proceeding to the processesing. This type of awareness is only available within
+the application.
+
+Applications can be modified to incorporate the extra contextual awareness in
+order to improve the reported busyness by marking areas of code as "busy" or
+"idle" appropriately. This can be done by inserting the timestamping macro::
+
+    RTE_LCORE_POLL_BUSYNESS_TIMESTAMP(0)    /* to mark section as idle */
+    RTE_LCORE_POLL_BUSYNESS_TIMESTAMP(32)   /* where 32 is nb_pkts to mark section as busy (non-zero is busy) */
+
+All cycles since the last state change (idle to busy, or vice versa) will be
+counted towards the current state's counter.
+
+Consuming the Telemetry
+-----------------------
+
+The telemetry gathered for lcore poll busyness can be read from the `telemetry.py`
+script via the new `/eal/lcore/poll_busyness` endpoint::
+
+    $ ./usertools/dpdk-telemetry.py
+    --> /eal/lcore/poll_busyness
+    {"/eal/lcore/poll_busyness": {"12": -1, "13": 85, "14": 84}}
+
+* Cores not collecting poll busyness will report "-1". E.g. control cores or inactive cores.
+* All enabled cores will report their poll busyness in the range 0-100.
+
+Enabling and Disabling Lcore Poll Busyness Telemetry
+----------------------------------------------------
+
+By default, the lcore poll busyness telemetry is disabled at compile time. In
+order to allow DPDK to gather this metric, the ``enable_lcore_poll_busyness``
+meson option must be set to ``true``.
+
+.. note::
+    Enabling lcore poll busyness telemetry may impact performance due to the
+    additional timestamping, potentially per poll depending on the application.
+    This can be measured with the `lcore_poll_busyness_perf_autotest`.
+
+At compile time
+^^^^^^^^^^^^^^^
+
+Support can be enabled/disabled at compile time via the meson option.
+It is disabled by default.::
+
+    $ meson configure -Denable_lcore_poll_busyness=true     #enable
+
+    $ meson configure -Denable_lcore_poll_busyness=false    #disable
+
+At run time
+^^^^^^^^^^^
+
+Support can also be enabled/disabled during runtime (if the meson option is
+enabled at compile time). Disabling at runtime comes at the cost of an additional
+branch, however no additional function calls are performed.
+
+To enable/disable support at runtime, a call can be made to the appropriately
+telemetry endpoint.
+
+Disable::
+
+    $ ./usertools/dpdk-telemetry.py
+    --> /eal/lcore/poll_busyness_disable
+    {"/eal/lcore/poll_busyness_disable": {"poll_busyness_enabled": 0}}
+
+Enable::
+
+    $ ./usertools/dpdk-telemetry.py
+    --> /eal/lcore/poll_busyness_enable
+    {"/eal/lcore/poll_busyness_enable": {"poll_busyness_enabled": 1}}