aboutsummaryrefslogtreecommitdiffstats
diff options
context:
space:
mode:
authorJan Gelety <jgelety@cisco.com>2020-09-15 01:09:40 +0200
committerTibor Frank <tifrank@cisco.com>2020-10-14 04:40:49 +0000
commit58b184079687bf4c834bba6cb32de9f6886c60b0 (patch)
treef24663f277a34ba39e17f0df578d2d08a7a55aa3
parent89c6334835128fd094238131abaf37cb3fd29117 (diff)
Report: vpp performance release notes update
Change-Id: I8c7e3af80a976a252a4fd4df223ab6122decc201 Signed-off-by: Jan Gelety <jgelety@cisco.com> (cherry picked from commit 99f4ec11d4daca69501b7f6aa4aeffa61f29062a)
Diffstat
-rw-r--r--docs/report/dpdk_performance_tests/csit_release_notes.rst18
-rw-r--r--docs/report/vpp_performance_tests/csit_release_notes.rst97
-rw-r--r--docs/report/vpp_performance_tests/overview.rst15
3 files changed, 64 insertions, 66 deletions
diff --git a/docs/report/dpdk_performance_tests/csit_release_notes.rst b/docs/report/dpdk_performance_tests/csit_release_notes.rst
index ceacd89eb3..829e7d2e27 100644
--- a/docs/report/dpdk_performance_tests/csit_release_notes.rst
+++ b/docs/report/dpdk_performance_tests/csit_release_notes.rst
@@ -6,14 +6,7 @@ Changes in |csit-release|
#. DPDK PERFORMANCE TESTS
- - Refactor of CSIT initialization helper scripts for DPDK testpmd
- and l3fwd. Removing obsolete parameters from DPDK testpmd and l3fwd
- command line.
-
- - Fixed 9000B L2 packet size tests not passing for all NICs.
-
- - Fixed DPDK compilation and test initialization of Mellanox NICs using
- CONFIG_RTE_LIBRTE_MLX5_PMD=y in compile configuration.
+ - Fixed DPDK compilation on ARM systems.
#. DPDK RELEASE VERSION CHANGE
@@ -29,6 +22,9 @@ List of known issues in |csit-release| for DPDK performance tests:
+----+------------------------------------------+----------------------------------------------------------------------------------------------------------+
| # | JiraID | Issue Description |
+====+==========================================+==========================================================================================================+
-| 1 | `CSIT-1701 | DPDK L3fwd tests with 9000B L2 packet size are not passing with Mellanox NICs. |
-| | <https://jira.fd.io/browse/CSIT-1701>`_ | L3fwd application does not accept parameter for increasing -mbuf-size in same way DPDK testpmd does. |
-+----+------------------------------------------+----------------------------------------------------------------------------------------------------------+
+| 1 | `CSIT-1761 | Denverton systems in FD.io CSIT lab (2n-dnv and 3n-dnv) reports dpdk compilation error very often. |
+| | <https://jira.fd.io/browse/CSIT-1761>`_ | |
++----+-----------------------------------------+-----------------------------------------------------------------------------------------------------------+
+| 2 | `CSIT-1762 | TRex reports link DOWN in case of dpdk testpmd tests on FD.io CSIT Denverton systems (2n-dnv and 3n-dnv). |
+| | <https://jira.fd.io/browse/CSIT-1762>`_ | |
++----+-----------------------------------------+-----------------------------------------------------------------------------------------------------------+
diff --git a/docs/report/vpp_performance_tests/csit_release_notes.rst b/docs/report/vpp_performance_tests/csit_release_notes.rst
index 1a66eab5eb..949419d393 100644
--- a/docs/report/vpp_performance_tests/csit_release_notes.rst
+++ b/docs/report/vpp_performance_tests/csit_release_notes.rst
@@ -10,14 +10,36 @@ Changes in |csit-release|
:ref:`test_environment_versioning`.
- To identify performance changes due to VPP code changes from
- v20.01.0 to v20.05.0, both have been tested in CSIT environment
- ver. 4 and compared against each other. All substantial
+ v20.05.0 to v20.09.0, both have been tested in CSIT environment
+ ver. 5 and compared against each other. All substantial
progressions has been marked up with RCA analysis. See
:ref:`vpp_compare_current_vs_previous_release` and
:ref:`vpp_known_issues`.
+ - **NAT44 tests**: Adapted existing and added new tests.
+
+ - Refactored NAT44 deterministic mode (nat44det) tests to use separate
+ det44 vpp plugin and to use the same scheme of inside and outside
+ addresses and ports, as used in new NAT44 endpoint-dependent mode tests.
+
+ - Added new NAT44 endpoint-depended mode uni-directional (nat44ed-udir)
+ tests that measure packet throughput in one direction with usage of TRex
+ in stateless mode.
+
+ - Added new NAT44 endpoint-dependent mode CPS tests that measure
+ connections per second with usage of TRex in stateful mode.
+ NOTE: CPS measurement data will be added in two weeks.
+
+ - **IPSec async mode tests**: Added VPP performance tests for async crypto
+ engine.
+
+ - **AMD 2n-zn2 testbed**: New physical testbed type installed in FD.io CSIT,
+ VPP performance data will be added in upcoming weeks.
+
#. TEST FRAMEWORK
+ - **TRex ASTF**: Added capability to run TRex in advanced stateful mode.
+
- **CSIT PAPI support**: Due to issues with PAPI performance, VAT is
still used in CSIT for all VPP scale tests. See known issues below.
@@ -25,6 +47,9 @@ Changes in |csit-release|
removal of redundant RF keywords and aligning of suite/test
setup/teardowns.
+ - **Intel E810CQ 100G NIC**: Added configuration for Intel E810CQ 100G NIC.
+ No tests run for this NIC as it is not present in FD.io CSIT lab yet.
+
#. PRESENTATION AND ANALYTICS LAYER
- **Graphs improvements**: Added possibility to use Gbps on Y-axis in
@@ -40,7 +65,7 @@ Changes in |csit-release|
Known Issues
------------
-List of known issues in |csit-release| for VPP performance tests:
+List of known issues in |csit-release| for VPP performance tests: TODO: check
+----+-----------------------------------------+-----------------------------------------------------------------------------------------------------------+
| # | JiraID | Issue Description |
@@ -51,23 +76,27 @@ List of known issues in |csit-release| for VPP performance tests:
| 2 | `VPP-662 | 9000B packets not supported by NICs VIC1227 and VIC1387. |
| | <https://jira.fd.io/browse/VPP-662>`_ | |
+----+-----------------------------------------+-----------------------------------------------------------------------------------------------------------+
-| 3 | `VPP-1677 | 9000B ip4 nat44: VPP crash + coredump. |
-| | <https://jira.fd.io/browse/VPP-1677>`_ | VPP crashes very often in case that NAT44 is configured and it has to process IP4 jumbo frames (9000B). |
+| 3 | `CSIT-1763 | Adapt ramp-up phase of nat44 tests for different frame sizes. |
+| | <https://jira.fd.io/browse/CSIT-1763>`_ | Currently ramp-up phase rate and duration values are correctly set for tests with 64B frame size. |
+----+-----------------------------------------+-----------------------------------------------------------------------------------------------------------+
-| 4 | `CSIT-1591 | All CSIT scale tests can not use PAPI due to much slower performance compared to VAT/CLI (it takes much |
-| | <https://jira.fd.io/browse/CSIT-1499>`_ | longer to program VPP). This needs to be addressed on the PAPI side. |
-| +-----------------------------------------+ |
-| | `VPP-1763 | |
-| | <https://jira.fd.io/browse/VPP-1763>`_ | |
+| 4 | `CSIT-1671 | All CSIT scale tests can not use PAPI due to much slower performance compared to VAT/CLI (it takes much |
+| | <https://jira.fd.io/browse/CSIT-1671>`_ | longer to program VPP). This needs to be addressed on the PAPI side. |
+| +-----------------------------------------+ The usual PAPI library spends too much time parsing arguments, so even with async processing (hundreds of |
+| | `VPP-1763 | commands in flight over socket), the VPP configuration for large scale tests (millions of messages) takes |
+| | <https://jira.fd.io/browse/VPP-1763>`_ | too long. |
+----+-----------------------------------------+-----------------------------------------------------------------------------------------------------------+
| 5 | `VPP-1675 | IPv4 IPSEC 9000B packet tests are failing as no packet is forwarded. |
| | <https://jira.fd.io/browse/VPP-1675>`_ | Reason: chained buffers are not supported. |
+----+-----------------------------------------+-----------------------------------------------------------------------------------------------------------+
-| 6 | `CSIT-1593 | IPv4 AVF 9000B packet tests are failing on 3n-skx while passing on 2n-skx. |
-| | <https://jira.fd.io/browse/CSIT-1593>`_ | |
+| 6 | `VPP-1934 | [i40e] Interfaces are not brought up from carrier-down. |
+| | <https://jira.fd.io/browse/VPP-1934 >`_ | In case of i40e -based interface (e.g Intel x700 series NIC) is bound to kernel driver (i40e) and is in |
+| | | state "no-carrier" (<NO-CARRIER,BROADCAST,MULTICAST,UP>) because previously it was disabled via |
+| | | "I40E_AQ_PHY_LINK_ENABLED" call, then VPP during initialization of AVF interface is not re-enabling |
+| | | interface link via i40e driver to up. |
+| | | CSIT implemented `workaround for AVF interface <https://gerrit.fd.io/r/c/csit/+/29086>`_ until fixed. |
+----+-----------------------------------------+-----------------------------------------------------------------------------------------------------------+
-| 7 | `CSIT-1679 | 2n-clx VPP ip4 tests with xxv710 and avf driver are sporadically failing. |
-| | <https://jira.fd.io/browse/CSIT-1679>`_ | |
+| 7 | `CSIT-1760 | All Mellanox / rdma driver tests are failing on LF testbed28 while successfully run on other LF testbeds. |
+| | <https://jira.fd.io/browse/CSIT-1760>`_ | |
+----+-----------------------------------------+-----------------------------------------------------------------------------------------------------------+
Root Cause Analysis for Performance Changes
@@ -78,42 +107,6 @@ List of RCAs in |csit-release| for VPP performance changes:
+----+-----------------------------------------+-----------------------------------------------------------------------------------------------------------+
| # | JiraID | Issue Description |
+====+=========================================+===========================================================================================================+
-| 1 | `CSIT-1731 | Confirm the cause for ip4scale -rnd regressions. |
-| | <https://jira.fd.io/browse/CSIT-1731>`_ | |
-+----+-----------------------------------------+-----------------------------------------------------------------------------------------------------------+
-| 2 | `CSIT-1735 | Identify cause of ethip6-ip6scale2m progression. |
-| | <https://jira.fd.io/browse/CSIT-1735>`_ | |
-+----+-----------------------------------------+-----------------------------------------------------------------------------------------------------------+
-| 3 | `CSIT-1738 | Identify cause for vppl2xc CSIT progressions. |
-| | <https://jira.fd.io/browse/CSIT-1738>`_ | |
-+----+-----------------------------------------+-----------------------------------------------------------------------------------------------------------+
-| 4 | `CSIT-1739 | Identify cause of ACL regressions. |
-| | <https://jira.fd.io/browse/CSIT-1739>`_ | |
-+----+-----------------------------------------+-----------------------------------------------------------------------------------------------------------+
-| 5 | `CSIT-1740 | Identify cause for avf-eth-l2xcbase CSIT progression. |
-| | <https://jira.fd.io/browse/CSIT-1740>`_ | |
-+----+-----------------------------------------+-----------------------------------------------------------------------------------------------------------+
-| 6 | `CSIT-1741 | Identify cause for vppl2xc VPP regressions. |
-| | <https://jira.fd.io/browse/CSIT-1741>`_ | |
-+----+-----------------------------------------+-----------------------------------------------------------------------------------------------------------+
-| 7 | `CSIT-1742 | Identify cause of ipsec CSIT regression. |
-| | <https://jira.fd.io/browse/CSIT-1742>`_ | |
-+----+-----------------------------------------+-----------------------------------------------------------------------------------------------------------+
-| 8 | `CSIT-1744 | Identify cause of memif VPP progression. |
-| | <https://jira.fd.io/browse/CSIT-1744>`_ | |
-+----+-----------------------------------------+-----------------------------------------------------------------------------------------------------------+
-| 9 | `CSIT-1745 | Verify cause of l2bdscale10kmaclrn VPP progression. |
-| | <https://jira.fd.io/browse/CSIT-1745>`_ | |
-+----+-----------------------------------------+-----------------------------------------------------------------------------------------------------------+
-| 10 | `CSIT-1746 | Identify cause for avf-dot1q-ip6base VPP progression. |
-| | <https://jira.fd.io/browse/CSIT-1746>`_ | |
-+----+-----------------------------------------+-----------------------------------------------------------------------------------------------------------+
-| 11 | `CSIT-1747 | Identify cause of ip4base-nat44 VPP progression. |
-| | <https://jira.fd.io/browse/CSIT-1747>`_ | |
-+----+-----------------------------------------+-----------------------------------------------------------------------------------------------------------+
-| 12 | `CSIT-1748 | Identify cause of aes128cbc-hmac512sha VPP progression. |
-| | <https://jira.fd.io/browse/CSIT-1748>`_ | |
-+----+-----------------------------------------+-----------------------------------------------------------------------------------------------------------+
-| 13 | `CSIT-1749 | Identify cause for l2bdbasemaclrn VPP progression in tests with dpdk app in VM. |
-| | <https://jira.fd.io/browse/CSIT-1749>`_ | |
+| 1 | | |
+| | | |
+----+-----------------------------------------+-----------------------------------------------------------------------------------------------------------+
diff --git a/docs/report/vpp_performance_tests/overview.rst b/docs/report/vpp_performance_tests/overview.rst
index 7fabebc05b..b316e81661 100644
--- a/docs/report/vpp_performance_tests/overview.rst
+++ b/docs/report/vpp_performance_tests/overview.rst
@@ -269,6 +269,10 @@ topologies and configurations:
size is set to the bi-directional link rate, unless there is a known
limitation preventing Traffic Generator from achieving the line rate.
+.. todo::
+
+ - Connections per second (CPS): TODO
+
|csit-release| includes following VPP data plane functionality
performance tested across a range of NIC drivers and NIC models:
@@ -292,6 +296,10 @@ performance tested across a range of NIC drivers and NIC models:
| IPv6 Scale | IPv6 routing with 20k, 200k and 2M FIB |
| | entries. |
+-----------------------+----------------------------------------------+
+| IPSecAsyncHW | IPSec encryption with AES-GCM, CBC-SHA-256 |
+| | ciphers in async mode, in combination with |
+| | IPv4 routing. Intel QAT HW acceleration. |
++-----------------------+----------------------------------------------+
| IPSecHW | IPSec encryption with AES-GCM, CBC-SHA-256 |
| | ciphers, in combination with IPv4 routing. |
| | Intel QAT HW acceleration. |
@@ -331,9 +339,10 @@ performance tested across a range of NIC drivers and NIC models:
| Memif | with different VPP forwarding modes incl. |
| | L2XC, L2BD. |
+-----------------------+----------------------------------------------+
-| NAT | (Source) Network Address Translation tests |
-| | with varying number of users and ports per |
-| | user. |
+| NAT44 | (Source) Network Address Translation |
+| | deterministic mode and endpoint-dependent |
+| | mode tests with varying number of users and |
+| | ports per user for IPv4. |
+-----------------------+----------------------------------------------+
| QoS Policer | Ingress packet rate measuring, marking and |
| | limiting (IPv4). |