Release Notes

Changes in CSIT-2206

1. VPP PERFORMANCE TESTS

   • 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.

2. TEST FRAMEWORK

   • Removed ASTF PPS tests: They provide no real benefit compared to TPUT tests. The ip4base variants renamed to TPUT.

   • Changed TCP TPUT profiles: The previous ones were found to be faulty. The new ones do not use bursts of packets to avoid CSIT-1830 and CSIT-1846.

   • CSIT test environment version has been updated to ver. 10, 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-1827	3n-icx, 3n-skx: all AVF crypto tests sporadically fail. 1518B with no traffic, IMIX with excessive packet loss.
2	CSIT-1830	All testbeds: All TCP tput (and pps) tests are failing for small packets.
3	CSIT-1832	3n-alt: NDR 1 packet lost on random tests.
4	CSIT-1834	2n-icx, 2n-skx: sporadic AVF soak tests failing to find critical load with PLRsearch.
5	CSIT-1846	2n-skx, 2n-clx, 2n-icx: ALL 1518B TCP tput tests failing with big packet loss.
6	CSIT-1847	2n-skx: all 10vm-1t test failed with half of packets dropped.
7	CSIT-1849	2n-skx: UDP 16m tput tests fail to create all sessions.

Previous

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

#	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-1802	AF-XDP - NDR tests failing from time to time.
4	CSIT-1803	3n-icx testbeds (Icelake): all IMIX aes128cbc-hmac512sha tests are failing due to excessive packet loss.
5	CSIT-1804	3n-tsh, 3n-alt testbed (Taishan, Altra): NDR tests failing from time to time.
6	CSIT-1812	All IMIX NAT44DET 4m 16m scale tests fail due to not creating required session count.
7	CSIT-1782	Multicore AVF tests are failing when trying to create interface. Frequency is reduced by CSIT workaround, but occasional failures do still happen.
8	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
9	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
10	CSIT-1801	9000B payload frames not forwarded over tunnels due to violating supported Max Frame Size (VxLAN, LISP, SRv6).
11	CSIT-1808	All tests with 9000B payload frames not forwarded over memif interfaces.
12	CSIT-1809	All tests with 9000B payload frames not forwarded over vhostuser interfaces.

Fixed

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

#	JiraID	Issue Description
1	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
2	CSIT-1811	All 9000B NAT44DET 64k 1m scale tests fail due to bps rate set to high on TRex.
3	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.

Root Cause Analysis for Performance Changes

List of RCAs in CSIT-2206 for VPP performance changes:

#	JiraID	Issue Description
1	CSIT-1851	trending regression: various icelake tests around 2202-04-15 Somewhat expected consequence of a VPP usability fix, the previous VPP compiler version was too new for the OS used.
2	VPP-2030	regression: ip6base on ICX around 2022-03-23 “Loads blocked due to overlapping with a preceding store that cannot be forwarded.” started happening in ip6-lookup graph node.
3	CSIT-1852	2n-zn2 mellanox performance cap Old issue, only now distinguished from CSIT-1751. This testbed+nic combination is capped below 28 Mpps, cause not identified yet.
4	CSIT-1853	trending regression: nat44ed cps around 2202-04-01 VPP change added more computation to slow path (in order to support multiple VRFs). Not clear if the VPP implementation is optimized enough.