Release Notes

Changes in CSIT-2101

  1. VPP PERFORMANCE TESTS

    • CSIT test environment is versioned, see Environment Versioning.

    • GENEVE tests: Added VPP performance tests for GENEVE tunnels. See Generic Network Virtualization Encapsulation for more details.

    • GSO tests: Added VPP performance tests for GSOtap and GSOvirtio. All tested topologies are compared with GSO enabled and disabled. In CSIT-2101 there is only 1t1c tests running. See Generic Segmentation Offload Tests for more details.

    • NAT44 tests: Added new test type, pure throughput tests. They are similar to PPS tests, but they employ ramp-up trials to ensure all sessions are created (and not timing out) for performance trials.

    • Jumbo for ipsec: Test cases with 9000 byte frames are re-enabled in ipsec suites.

    • Randomized profiles: Improved repeatability and cycle length. For details, see Packet Flow Ordering.

    • Arm 2n-tx2 testbed: New physical testbed type installed in FD.io CSIT, with VPP and DPDK performance data added to CSIT trending and this report.

    • Framework speedup: Shortened overall test job duration by using a different test selection mechanism (using –test instead of –include) and by avoiding unnecessary PAPI reconnects.

  2. TEST FRAMEWORK

    • TRex ASTF: Improved 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.

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

  3. PRESENTATION AND ANALYTICS LAYER

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

Known Issues

List of known issues in CSIT-2101 for VPP performance tests:

#

JiraID

Issue Description

1

CSIT-570

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

VPP-662

9000B packets not supported by NICs VIC1227 and VIC1387.

3

CSIT-1763

Adapt ramp-up phase of nat44 tests for different frame sizes. Currently ramp-up phase rate and duration values are correctly set for tests with 64B frame size.

4

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. The usual PAPI library spends too much time parsing arguments, so even with async processing (hundreds of commands in flight over socket), the VPP configuration for large scale tests (millions of messages) takes too long.

VPP-1763

5

VPP-1934

[i40e] Interfaces are not brought up from carrier-down. 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 until fixed.

Root Cause Analysis for Performance Changes

List of RCAs in CSIT-2101 for VPP performance changes:

#

JiraID

Issue Description

1

VPP-1972 <https://jira.fd.io/browse/VPP-1972>

One VPP change has decreased performance of NAT44ed processing, both slow path and fast path.