Release Notes¶
Changes in CSIT-2001¶
VPP PERFORMANCE TESTS
Intel Xeon 2n-skx, 3n-skx testbeds: VPP performance test data is now included in this report version. Due to substantial impact of test environment changes (applied during the CSIT-2001 development cycle) on the performance of VPP software, a new approach to performance comparison and progression/regression root cause analysis (RCA) has been applied.
CSIT test environment is now versioned, with ver. 1 associated with CSIT rls1908 git branch as of 2019-08-21, and ver. 2 associated with CSIT rls2001 git branch as of 2020-03-27.
To identify performance changes due to CSIT test environment changes from ver. 1 to ver. 2, VPP v19.08.1 has been re-tested in ver. 2 and compared against the past results obtained with testing in ver. 1. Separate RCA1 analysis has been applied to this part. See Current vs. Previous Release and Known Issues.
To identify performance changes due to VPP code changes from v19.08.1 to v20.01.0, both have been tested in CSIT environment ver. 2 and compared against each other. Separate RCA2 analysis has been applied to this part. See Current vs. Previous Release and Known Issues.
Intel Xeon 2n-clx testbeds: VPP performance test data is now included in this report. See Known Issues.
Service density 2n-skx tests: Added new NF density tests with IPsec encryption between DUTs.
AVF tests: Full test coveraged based on code changes in CSIT core layer (driver/interface awareness) and generated by suite generator (Intel Fortville NICs only).
Hoststack tests: Major refactor of VPP Hoststack TCP/IP performance tests using WRK generator talking to the VPP HTTP static server plugin measuring connections per second and requests per second. Added new iperf3 with LDPreload tests, iperf3/LDPreload tests with packet loss induced via the VPP NSIM (Network Simulator) plugin, and QUIC/UDP/IP transport tests. All of the new tests measure goodput through the VPP Hoststack from client to server.
Latency HDRHistogram: Added High Dynamic Range Histogram latency measurements based on the new capability in TRex traffic generator. HDRH latency data presented in latency packet percentile graphs and in detailed results tables.
Mellanox CX556A-EDAT tests: Added tests with Mellanox ConnectX5-2p100GE NICs in 2n-clx testbeds using VPP native rdma driver.
IPsec reconfiguration tests: Added tests measuring the impact of IPsec tunnels creations and removals.
Load Balancer tests: Added VPP performance tests for Maglev, L3DSR (Direct Server Return), Layer 4 Load Balancing NAT Mode.
TEST FRAMEWORK
CSIT Python3 support: Full migration of CSIT from Python2.7 to Python3.6. This change includes library migration, PIP dependency upgrade, CSIT container images, infrastructure packages ugrade/installation.
CSIT PAPI support: Finished conversion of CSIT VAT L1 keywords to PAPI L1 KWs in CSIT using VPP Python bindings (VPP PAPI). Redesign of key components of PAPI Socket Executor and PAPI history. Due to issues with PAPI performance, VAT is still used in CSIT for all VPP scale tests. See known issues below.
Test Suite Generator: Added capability to generate suites for different drivers per NIC model including DPDK, AVF, RDMA. Extended coverage for all tests.
General Code Housekeeping: Ongoing RF keywords optimizations, removal of redundant RF keywords and aligning of suite/test setup/teardowns.
TEST ENVIRONMENT
TRex Fortville NIC Performance: Received FVL fix from Intel resolving TRex low throughput issue. TRex per FVL NIC throughput increased from ~27 Mpps to the nominal ~37 Mpps. For detail see CSIT-1503 and TRex-519].
New Intel Xeon Cascadelake Testbeds: Added performance tests for 2-Node-Cascadelake (2n-clx) testbeds with x710, xxv710 and cx556a-edat NIC cards.
PRESENTATION AND ANALYTICS LAYER
Graphs layout improvements: Improved performance graphs layout for better readibility and maintenance: test grouping, axis labels, descriptions, other informative decoration.
Latency graphs: Min/Avg/Max group bar latency graphs are replaced with packet latency percentile distributon at different background packet loads based on TRex latency hdrhistogram measurements.
Known Issues¶
List of known issues in CSIT-2001 for VPP performance tests:
# |
JiraID |
Issue Description |
|---|---|---|
1 |
Sporadic (1 in 200) NDR discovery test failures on x520. DPDK reporting rx-errors, indicating L1 issue. Suspected issue with HW combination of X710-X520 in LF testbeds. Not observed outside of LF testbeds. |
|
2 |
9000B packets not supported by NICs VIC1227 and VIC1387. |
|
3 |
Memif tests are sporadically failing on initialization of memif connection. |
|
4 |
9000B ip4 nat44: VPP crash + coredump. VPP crashes very often in case that NAT44 is configured and it has to process IP4 jumbo frames (9000B). |
|
5 |
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. |
|
6 |
IPv4 IPSEC 9000B packet tests are failing as no packet is forwarded. Reason: chained buffers are not supported. |
|
7 |
IPv4 AVF 9000B packet tests are failing on 3n-skx while passing on 2n-skx. |
|
8 |
Intel Xeon 2n-skx, 3n-skx and 2n-clx testbeds behaviour and performance became inconsistent following the upgrade to the latest Ubuntu 18.04 LTS kernel version (4.15.0-72-generic) and associated microcode packages (skx ucode 0x2000064, clx ucode 0x500002c). VPP as well as DPDK L3fwd tests are affected. |
|
9 |
All 2n-clx VPP ip4 tests with xxv710 and avf driver are failing. |
|
10 |
Some 2n-clx cx556a rdma tests are failing. |
|
11 |
Root Cause Analysis for CSIT-2001. Investigate high stdev of tests with VPP inside VM. |
|
12 |
Root Cause Analysis for CSIT-2001. Identify cause of dot1q-l2xcbase progression. |
|
13 |
Root Cause Analysis for CSIT-2001. Identify cause of avf-ip4scale regression. |
|
14 |
Root Cause Analysis for CSIT-2001. Identify cause of progression in vhost-user tests with testpmd in VM. |
|
15 |
Root Cause Analysis for CSIT-2001. Identify cause for avf-ip4base regression. |
|