Release Notes

Changes in CSIT-2202

1. VPP PERFORMANCE TESTS

   • Regressions with DPDK drivers: Change from DPDK v21.08 to DPDK v21.11 introduced regression across all tests using dpdk drivers (with dpdk_plugin loaded). Compared to previous VPP release performance drop varies in the range of -15% to -6%, depending on test. It is related to updated MTU checks within DPDK code and associated VPP code changes. See VPP v2202 release notes and Known Issues.

   • Number of CSIT 9000B frame tests failing: tests with higher encapsulation overhead are failing due to exceeding default Ethernet Maximum Frame Size value that has been reduced by MTU related VPP code changes. See VPP v2202 release notes and Known Issues.

   • Intel Xeon Ice Lake: Performance test data for these platforms is now provided by testbeds newly installed in FD.io CSIT labs. For details about the physical setup see 2-Node Xeon Icelake (2n-icx) and 3-Node Xeon Icelake (3n-icx).

   • Reduction of tests: Removed certain test variations executed iteratively for the report (as well as in daily and weekly trending) due to physical testbeds overload.

   • RC2 coverage test data is used for 2n-icx test bed: There is only one 2n-icx test bed and the amount of tests is large (same as 2n-skx with 4 test beds), so we decided to use test data already available from RC2 testing.

2. TEST FRAMEWORK

   • CSIT test environment version has been updated to ver. 9, see Environment Versioning.

   • CSIT PAPI support: Due to issues with PAPI performance, and deprecation of VAT, VPP CLI is used in CSIT for many VPP scale tests. See Known Issues.

   • General Code Housekeeping: Ongoing code optimizations and bug fixes.

3. PRESENTATION AND ANALYTICS LAYER

   • Graphs improvements: Updated Packet Latency graphs, see Packet Latency.

Known Issues

New

#	JiraID	Issue Description
1	CSIT-1799	All NAT44-ED 16M scale tests fail while setting NAT44 address range.
2	CSIT-1800	All Geneve L3 mode scale tests (1024 tunnels) are failing.
3	CSIT-1801	9000B payload frames not forwarded over tunnels due to violating supported Max Frame Size (VxLAN, LISP, SRv6).
4	CSIT-1808	All tests with 9000B payload frames not forwarded over memif interfaces.
5	CSIT-1809	All tests with 9000B payload frames not forwarded over vhostuser interfaces.
6	CSIT-1802	AF-XDP - NDR tests failing from time to time.
7	CSIT-1803	3n-icx testbeds (Icelake): all IMIX aes128cbc-hmac512sha tests are failing due to excessive packet loss.
8	CSIT-1804	3n-tsh testbed (Taishan): NDR tests failing from time to time.
9	CSIT-1805	3n-skx testbeds (Skylake): all hoststack quic vppecho scale tests are failing.
10	CSIT-1810	DPDK - performance regression with DPDK driver when Max Frame Size is set to less than 2023. Worse performance with DPDK driver when MTU is set to 2022 or less.
	VPP-1876
11	CSIT-1811	All 9000B NAT44DET 64k 1m scale tests fail due to bps rate set to high on TRex.
12	CSIT-1812	All IMIX NAT44DET 4m 16m scale tests fail due to not creating required session count.

Previous

Issues reported in previous releases which still affect the current results.

#	JiraID	Issue Description
1	CSIT-1782	Multicore AVF tests are failing when trying to create interface. Frequency is reduced by CSIT workaround, but occasional failures do still happen.
2	CSIT-1671	All CSIT scale tests can not use PAPI due to much slower performance compared to VAT/CLI (it takes much longer to program VPP). This needs to be addressed on the PAPI side. Currently, the time critical code uses VAT running large files with exec statements and CLI commands. Still, we needed to reduce the number of scale tests run to keep overall duration reasonable. More improvements needed to achieve sufficient configuration speed.
	VPP-1763
3	CSIT-1785	NAT44ED tests failing to establish all TCP sessions. At least for max scale, in allotted time (limited by session 500s timeout) due to worse slow path performance than previously measured and calibrated for. CSIT removed the max scale NAT tests to avoid this issue.
	VPP-1972
4	CSIT-1791	Performance regression in RDMA tests, due to CSIT environment changes. Two symptoms: 1. 10-20% regression across most tests. 2. DUT performance cap just below 38 Mpps.

Fixed

Issues reported in previous releases which were fixed in this release:

#	JiraID	Issue Description
1	CSIT-1789	IPSEC SW async scheduler tests show bad behavior. VPP code is not behaving correctly when crypto workers are the bottleneck.
	VPP-1998

Root Cause Analysis for Performance Changes

List of RCAs in CSIT-2202 for VPP performance changes:

#	JiraID	Issue Description
1	CSIT-1814	rls2202 regression: NAT44ED CPS larger scales.
2	CSIT-1815	rls2202 regression: l2bdbasemaclrn around 2021-11-11.
3	CSIT-1816	rls2202 regression: l2-input.
4	CSIT-1820	rls2202 regression: tx2 ip6base at dpdk upgrade.
5	CSIT-1821	rls2202 regression: tx2 ip6base before dpdk upgrade.
6	CSIT-1822	rls2202 regression: avf-ip4base around 2022-01-06.
7	CSIT-1823	rls2202 regression: vxlan-l2xcbase on tsh around 2022-01-07.
8	CSIT-1824	rls2202 regression: SRv6.