From 556932e2cf1143237e0cfc8b3acc8d17bbf49777 Mon Sep 17 00:00:00 2001 From: Maciek Konstantynowicz Date: Wed, 19 Jul 2017 11:20:44 +0100 Subject: rls1707 report: edits of vpp_performance_tests/rls_notes Change-Id: Ia7fbe0fbfa83c7f61f59850ea98e00405fc82711 Signed-off-by: Maciek Konstantynowicz (cherry picked from commit 912acb80b6607a6b904242a0ec695c874580bd02) --- .../vpp_performance_tests/csit_release_notes.rst | 108 ++++++++++++++++----- 1 file changed, 82 insertions(+), 26 deletions(-) diff --git a/docs/report/vpp_performance_tests/csit_release_notes.rst b/docs/report/vpp_performance_tests/csit_release_notes.rst index 0a5d25e6b6..782fa42840 100644 --- a/docs/report/vpp_performance_tests/csit_release_notes.rst +++ b/docs/report/vpp_performance_tests/csit_release_notes.rst @@ -4,45 +4,95 @@ CSIT Release Notes Changes in CSIT |release| ------------------------- -#. VPP performance test environment changes +#. Test environment changes in VPP data plane performance tests: - - Further optimizations of VM and vhost-user test environment - various - Qemu virtio queue size testing with value of 256 and 1024. Applied - Linux CFS optimization to run VPP worker threads and Qemu worker threads - with highest priority. + - Further characterization and optimizations of VPP vhost-user and VM + test methodology and test environment; -#. VPP performance test framework changes + - Tests with varying Qemu virtio queue (a.k.a. vring) sizes: + [vr256] default 256 descriptors, [vr1024] 1024 descriptors to + optimize for packet throughput; - - Full code review, optimization and refactor. + - Tests with varying Linux CFS (Completely Fair Scheduler) + settings: [cfs] default settings, [cfsrr1] CFS RoundRobin(1) + policy applied to all data plane threads handling test packet + path including all VPP worker threads and all Qemu testpmd + poll-mode threads; -#. T-rex changes + - Resulting test cases are all combinations with [vr256,vr1024] and + [cfs,cfsrr1] settings; - - Full refactor of T-rex driver and introduce of traffic profiles that - improves readability, manageability of traffic profiles for various - test scenarios. + - For more detail see performance results observations section in + this report; + +#. Code updates and optimizations in CSIT performance framework: + + - Complete CSIT framework code revision and optimizations as descried + on CSIT wiki page + `Design_Optimizations `_. + + - For more detail see the CSIT Framework Design section in this + report; + +#. Changes to CSIT driver for TRex Traffic Generator: + + - Complete refactor of TRex CSIT driver; + + - Introduction of packet traffic profiles to improve usability and + manageability of traffic profiles for a growing number of test + scenarios. + + - Support for packet traffic profiles to test IPv4/IPv6 stateful and + stateless DUT data plane features; #. Added VPP performance tests - - **LXC memif** + - **Linux Container VPP memif virtual interface tests** - - Memif interface tests interconnecting two VPP instances on single SUT. - Master VPP instance running on native OS with Intel x520 NIC and guest - VPP instance running in Linux Container (LXC) doing the L2 cross - connect loop. LXC running in privileged mode is pinned to dedicated - cores. All VPP instances are same version. + - VPP Memif virtual interface (shared memory interface) tests + interconnecting VPP instances over memif. VPP vswitch + instance runs in bare-metal user-mode handling Intel x520 NIC + 10GbE interfaces and connecting over memif (Master side) virtual + interfaces to another instance of VPP running in bare-metal Linux + Container (LXC) with memif virtual interfaces (Slave side). LXC + runs in a priviliged mode with VPP data plane worker threads + pinned to dedicated physical CPU cores per usual CSIT practice. + Both VPP run the same version of software. This test topology is + equivalent to existing tests with vhost-user and VMs. - **Stateful Security Groups** - - **VM vhost use cases** + - New tests of VPP stateful security-groups a.k.a. acl-plugin + functionally compatible with networking-vpp OpenStack; + + - New tested security-groups access-control-lists (acl) + configuration variants include: [iaclNsl] input acl stateless, + [oaclNsl] output acl stateless, [iaclNsf] input acl stateful + a.k.a. reflect, [oaclNsf] output acl stateful a.k.a. reflect, + where N is number of access-control-entries (ace) in the acl. + + - Testing packet flows transmitted by TG: 100, 10k, 100k, always + hitting the last permit entry in acl. + + - **VPP vhost and VM tests** + + - New VPP vhost-user and VM test cases to benchmark performance of + VPP and VM topologies with Qemu and CFS policy combinations of + [vr256,vr1024] x [cfs,cfsrr1]; + + - Statistical analysis of repeatibility of results; Performance Improvements ------------------------ Substantial improvements in measured packet throughput have been observed in a number of CSIT |release| tests listed below, with relative -increase of double-digit percentage points. Relative improvements are -calculated against the test results listed in CSIT |release-1| report. -VPP-16.09 numbers are provided for reference. +increase of double-digit percentage points. Relative improvements for +this release are calculated against the test results listed in CSIT +|release-1| report. The comparison is calculated between the mean values +based on collected and archived test results' samples for involved VPP +releases. Standard deviation has been also listed for CSIT |release|. +VPP-16.09 and VPP-17.01 numbers are provided for reference. NDR Throughput ~~~~~~~~~~~~~~ @@ -105,21 +155,27 @@ Here is the list of known issues in CSIT |release| for VPP performance tests: +---+-------------------------------------------------+------------+-----------------------------------------------------------------+ | # | Issue | Jira ID | Description | +---+-------------------------------------------------+------------+-----------------------------------------------------------------+ -| 1 | NDR discovery test failures 1518B frame size | VPP-663 | VPP reporting errors: dpdk-input Rx ip checksum errors. | +| 1 | Security-groups acl-plugin scale tests failure | CSIT-xxx | VPP with 2 worker threads crashes during security-groups | +| | with stateful acls if VPP with 2 worker threads | VPP-912 | iaclNsf and oaclNsf tests with 100k flows. | ++---+-------------------------------------------------+------------+-----------------------------------------------------------------+ +| 2 | VPP fails memif tests in 4 worker 2 core setup | CSIT-xxx | VPP with 4 worker threads running on 2 physical cores crashes | +| | | VPP-xxx | during memif tests. Initial debugging points to DPDK code | ++---+-------------------------------------------------+------------+-----------------------------------------------------------------+ +| X | NDR discovery test failures 1518B frame size | VPP-663 | VPP reporting errors: dpdk-input Rx ip checksum errors. | | | for ip4scale200k, ip4scale2m scale IPv4 routed- | | Observed frequency: all test runs. | | | forwarding tests. ip4scale20k tests are fine. | | | +---+-------------------------------------------------+------------+-----------------------------------------------------------------+ -| 2 | Vic1385 and Vic1227 low performance. | VPP-664 | Low NDR performance. | +| X | Vic1385 and Vic1227 low performance. | VPP-664 | Low NDR performance. | | | | | | +---+-------------------------------------------------+------------+-----------------------------------------------------------------+ -| 3 | Sporadic NDR discovery test failures on x520. | CSIT-750 | Suspected issue with HW settings (BIOS, FW) in LF | +| X | Sporadic NDR discovery test failures on x520. | CSIT-750 | Suspected issue with HW settings (BIOS, FW) in LF | | | | | infrastructure. Issue can't be replicated outside LF. | +---+-------------------------------------------------+------------+-----------------------------------------------------------------+ -| 4 | VPP in 2t2c setups - large variation | CSIT-568 | Suspected NIC firmware or DPDK driver issue affecting NDR | +| X | VPP in 2t2c setups - large variation | CSIT-568 | Suspected NIC firmware or DPDK driver issue affecting NDR | | | of discovered NDR throughput values across | | throughput. Applies to XL710 and X710 NICs, x520 NICs are fine. | | | multiple test runs with xl710 and x710 NICs. | | | +---+-------------------------------------------------+------------+-----------------------------------------------------------------+ -| 5 | Lower than expected NDR and PDR throughput with | CSIT-569 | Suspected NIC firmware or DPDK driver issue affecting NDR and | +| X | Lower than expected NDR and PDR throughput with | CSIT-569 | Suspected NIC firmware or DPDK driver issue affecting NDR and | | | xl710 and x710 NICs, compared to x520 NICs. | | PDR throughput. Applies to XL710 and X710 NICs. | +---+-------------------------------------------------+------------+-----------------------------------------------------------------+ -- cgit 1.2.3-korg