2.1. Overview

2.1.1. Tested Physical Topologies

CSIT VPP performance tests are executed on physical baremetal servers hosted by LF FD.io project. Testbed physical topology is shown in the figure below.:

+------------------------+           +------------------------+
|                        |           |                        |
|  +------------------+  |           |  +------------------+  |
|  |                  |  |           |  |                  |  |
|  |                  <----------------->                  |  |
|  |       DUT1       |  |           |  |       DUT2       |  |
|  +--^---------------+  |           |  +---------------^--+  |
|     |                  |           |                  |     |
|     |            SUT1  |           |  SUT2            |     |
+------------------------+           +------------------^-----+
      |                                                 |
      |                                                 |
      |                  +-----------+                  |
      |                  |           |                  |
      +------------------>    TG     <------------------+
                         |           |
                         +-----------+

SUT1 and SUT2 are two System Under Test servers (Cisco UCS C240, each with two Intel XEON CPUs), TG is a Traffic Generator (TG, another Cisco UCS C240, with two Intel XEON CPUs). SUTs run VPP SW application in Linux user-mode as a Device Under Test (DUT). TG runs TRex SW application as a packet Traffic Generator. Physical connectivity between SUTs and to TG is provided using different NIC models that need to be tested for performance. Currently installed and tested NIC models include:

  1. 2port10GE X520-DA2 Intel.
  2. 2port10GE X710 Intel.
  3. 2port10GE VIC1227 Cisco.
  4. 2port40GE VIC1385 Cisco.
  5. 2port40GE XL710 Intel.

From SUT and DUT perspective, all performance tests involve forwarding packets between two physical Ethernet ports (10GE or 40GE). Due to the number of listed NIC models tested and available PCI slot capacity in SUT servers, in all of the above cases both physical ports are located on the same NIC. In some test cases this results in measured packet throughput being limited not by VPP DUT but by either the physical interface or the NIC capacity.

Going forward CSIT project will be looking to add more hardware into FD.io performance labs to address larger scale multi-interface and multi-NIC performance testing scenarios.

For service chain topology test cases that require DUT (VPP) to communicate with VirtualMachines (VMs) or with Linux/Docker Containers (Ctrs) over vhost-user/memif interfaces, N of VM/Ctr instances are created on SUT1 and SUT2. Three types of service chain topologies are tested in CSIT rls1801:

  1. “Parallel” topology with packets flowing from NIC via DUT (VPP) to VM/Container and back to VPP and NIC;
  2. “Chained” topology (a.k.a. “Snake”) with packets flowing via DUT (VPP) to VM/Container, back to DUT, then to the next VM/Container, back to DUT and so on until the last VM/Container in a chain, then back to DUT and NIC;
  3. “Horizontal” topology with packets flowing via DUT (VPP) to Container, then via “horizontal” memif to the next Container, and so on until the last Container, then back to DUT and NIC. “Horizontal” topology is not supported for VMs;

For each of the above topologies, DUT (VPP) is tested in a range of L2 or IPv4/IPv6 configurations depending on the test suite. A sample DUT “Chained” service topology with N of VM/Ctr instances is shown in the figure below. Packet flow thru the DUTs and VMs/Ctrs is marked with ***:

+-------------------------+           +-------------------------+
| +---------+ +---------+ |           | +---------+ +---------+ |
| |VM/Ctr[1]| |VM/Ctr[N]| |           | |VM/Ctr[1]| |VM/Ctr[N]| |
| |  *****  | |  *****  | |           | |  *****  | |  *****  | |
| +--^---^--+ +--^---^--+ |           | +--^---^--+ +--^---^--+ |
|   *|   |*     *|   |*   |           |   *|   |*     *|   |*   |
| +--v---v-------v---v--+ |           | +--v---v-------v---v--+ |
| |  *   *       *   *  |*|***********|*|  *   *       *   *  | |
| |  *   *********   ***<-|-----------|->***   *********   *  | |
| |  *    DUT1          | |           | |       DUT2       *  | |
| +--^------------------+ |           | +------------------^--+ |
|   *|                    |           |                    |*   |
|   *|            SUT1    |           |  SUT2              |*   |
+-------------------------+           +-------------------------+
    *|                                                     |*
    *|                                                     |*
    *|                    +-----------+                    |*
    *|                    |           |                    |*
    *+-------------------->    TG     <--------------------+*
    **********************|           |**********************
                          +-----------+

In above “Chained” topology, packets are switched by DUT multiple times: twice for a single VM/Ctr, three times for two VMs/Ctrs, N+1 times for N VMs/Ctrs. Hence the external throughput rates measured by TG and listed in this report must be multiplied by (N+1) to represent the actual DUT aggregate packet forwarding rate.

For a “Parallel” and “Horizontal” service topologies packets are always switched by DUT twice per service chain.

Note that reported DUT (VPP) performance results are specific to the SUTs tested. Current LF FD.io SUTs are based on Intel XEON E5-2699v3 2.3GHz CPUs. SUTs with other CPUs are likely to yield different results. A good rule of thumb, that can be applied to estimate VPP packet thoughput for Phy-to-Phy (NIC-to-NIC, PCI-to-PCI) topology, is to expect the forwarding performance to be proportional to CPU core frequency, assuming CPU is the only limiting factor and all other SUT parameters equivalent to FD.io CSIT environment. The same rule of thumb can be also applied for Phy-to-VM/Ctr-to-Phy (NIC-to-VM/Ctr-to-NIC) topology, but due to much higher dependency on intensive memory operations and sensitivity to Linux kernel scheduler settings and behaviour, this estimation may not always yield good enough accuracy.

For detailed FD.io CSIT testbed specification and topology, as well as configuration and setup of SUTs and DUTs testbeds please refer to Test Environment.

Similar SUT compute node and DUT VPP settings can be arrived to in a standalone VPP setup by using a vpp-config configuration tool developed within the VPP project using CSIT recommended settings and scripts.

2.1.2. Performance Tests Coverage

Performance tests are split into two main categories:

  • Throughput discovery - discovery of packet forwarding rate using binary search in accordance to RFC 2544.
    • NDR - discovery of Non Drop Rate packet throughput, at zero packet loss; followed by one-way packet latency measurements at 10%, 50% and 100% of discovered NDR throughput.
    • PDR - discovery of Partial Drop Rate, with specified non-zero packet loss currently set to 0.5%; followed by one-way packet latency measurements at 100% of discovered PDR throughput.
  • Throughput verification - verification of packet forwarding rate against previously discovered throughput rate. These tests are currently done against 0.9 of reference NDR, with reference rates updated periodically.

CSIT rls1801 includes following performance test suites, listed per NIC type:

  • 2port10GE X520-DA2 Intel
    • L2XC - L2 Cross-Connect switched-forwarding of untagged, dot1q, dot1ad VLAN tagged Ethernet frames.
    • L2BD - L2 Bridge-Domain switched-forwarding of untagged Ethernet frames with MAC learning; disabled MAC learning i.e. static MAC tests to be added.
    • L2BD Scale - L2 Bridge-Domain switched-forwarding of untagged Ethernet frames with MAC learning; disabled MAC learning i.e. static MAC tests to be added with 20k, 200k and 2M FIB entries.
    • IPv4 - IPv4 routed-forwarding.
    • IPv6 - IPv6 routed-forwarding.
    • IPv4 Scale - IPv4 routed-forwarding with 20k, 200k and 2M FIB entries.
    • IPv6 Scale - IPv6 routed-forwarding with 20k, 200k and 2M FIB entries.
    • VMs with vhost-user - virtual topologies with 1 VM and service chains of 2 VMs using vhost-user interfaces, with VPP forwarding modes incl. L2 Cross-Connect, L2 Bridge-Domain, VXLAN with L2BD, IPv4 routed-forwarding.
    • COP - IPv4 and IPv6 routed-forwarding with COP address security.
    • ACL - L2 Bridge-Domain switched-forwarding and IPv4 and IPv6 routed- forwarding with iACL and oACL IP address, MAC address and L4 port security.
    • LISP - LISP overlay tunneling for IPv4-over-IPv4, IPv6-over-IPv4, IPv6-over-IPv6, IPv4-over-IPv6 in IPv4 and IPv6 routed-forwarding modes.
    • VXLAN - VXLAN overlay tunnelling integration with L2XC and L2BD.
    • QoS Policer - ingress packet rate measuring, marking and limiting (IPv4).
    • NAT - (Source) Network Address Translation tests with varying number of users and ports per user.
    • Container memif connections - VPP memif virtual interface tests to interconnect VPP instances with L2XC and L2BD.
    • Container K8s Orchestrated Topologies - Container topologies connected over the memif virtual interface.
    • SRv6 - Segment Routing IPv6 tests.
  • 2port40GE XL710 Intel
    • L2XC - L2 Cross-Connect switched-forwarding of untagged Ethernet frames.
    • L2BD - L2 Bridge-Domain switched-forwarding of untagged Ethernet frames with MAC learning.
    • IPv4 - IPv4 routed-forwarding.
    • IPv6 - IPv6 routed-forwarding.
    • VMs with vhost-user - virtual topologies with 1 VM and service chains of 2 VMs using vhost-user interfaces, with VPP forwarding modes incl. L2 Cross-Connect, L2 Bridge-Domain, VXLAN with L2BD, IPv4 routed-forwarding.
    • IPSecSW - IPSec encryption with AES-GCM, CBC-SHA1 ciphers, in combination with IPv4 routed-forwarding.
    • IPSecHW - IPSec encryption with AES-GCM, CBC-SHA1 ciphers, in combination with IPv4 routed-forwarding. Intel QAT HW acceleration.
    • IPSec+LISP - IPSec encryption with CBC-SHA1 ciphers, in combination with LISP-GPE overlay tunneling for IPv4-over-IPv4.
    • VPP TCP/IP stack - tests of VPP TCP/IP stack used with VPP built-in HTTP server.
  • 2port10GE X710 Intel
    • L2BD - L2 Bridge-Domain switched-forwarding of untagged Ethernet frames with MAC learning.
    • VMs with vhost-user - virtual topologies with 1 VM using vhost-user interfaces, with VPP forwarding modes incl. L2 Bridge-Domain.
  • 2port10GE VIC1227 Cisco
    • L2BD - L2 Bridge-Domain switched-forwarding of untagged Ethernet frames with MAC learning.
  • 2port40GE VIC1385 Cisco
    • L2BD - L2 Bridge-Domain switched-forwarding of untagged Ethernet frames
      with MAC learning.

Execution of performance tests takes time, especially the throughput discovery tests. Due to limited HW testbed resources available within FD.io labs hosted by LF, the number of tests for NICs other than X520 (a.k.a. Niantic) has been limited to few baseline tests. CSIT team expect the HW testbed resources to grow over time, so that complete set of performance tests can be regularly and(or) continuously executed against all models of hardware present in FD.io labs.

2.1.3. Performance Tests Naming

CSIT rls1801 follows a common structured naming convention for all performance and system functional tests, introduced in CSIT rls1710.

The naming should be intuitive for majority of the tests. Complete description of CSIT test naming convention is provided on CSIT test naming wiki.

2.1.4. Methodology: Multi-Core and Multi-Threading

Intel Hyper-Threading - CSIT rls1801 performance tests are executed with SUT servers’ Intel XEON processors configured in Intel Hyper-Threading Disabled mode (BIOS setting). This is the simplest configuration used to establish baseline single-thread single-core application packet processing and forwarding performance. Subsequent releases of CSIT will add performance tests with Intel Hyper-Threading Enabled (requires BIOS settings change and hard reboot of server).

Multi-core Tests - CSIT rls1801 multi-core tests are executed in the following VPP thread and core configurations:

  1. 1t1c - 1 VPP worker thread on 1 CPU physical core.
  2. 2t2c - 2 VPP worker threads on 2 CPU physical cores.
  3. 4t4c - 4 VPP worker threads on 4 CPU physical cores.

VPP worker threads are the data plane threads. VPP control thread is running on a separate non-isolated core together with other Linux processes. Note that in quite a few test cases running VPP workers on 2 or 4 physical cores hits the I/O bandwidth or packets-per-second limit of tested NIC.

Section throughput_speedup_multi_core includes a set of graphs illustrating packet throughout speedup when running VPP on multiple cores.

2.1.5. Methodology: Packet Throughput

Following values are measured and reported for packet throughput tests:

  • NDR binary search per RFC 2544:
    • Packet rate: “RATE: <aggregate packet rate in packets-per-second> pps (2x <per direction packets-per-second>)”;
    • Aggregate bandwidth: “BANDWIDTH: <aggregate bandwidth in Gigabits per second> Gbps (untagged)”;
  • PDR binary search per RFC 2544:
    • Packet rate: “RATE: <aggregate packet rate in packets-per-second> pps (2x <per direction packets-per-second>)”;
    • Aggregate bandwidth: “BANDWIDTH: <aggregate bandwidth in Gigabits per second> Gbps (untagged)”;
    • Packet loss tolerance: “LOSS_ACCEPTANCE <accepted percentage of packets lost at PDR rate>”;
  • NDR and PDR are measured for the following L2 frame sizes:
    • IPv4: 64B, IMIX_v4_1 (28x64B,16x570B,4x1518B), 1518B, 9000B;
    • IPv6: 78B, 1518B, 9000B;
  • NDR and PDR binary search resolution is determined by the final value of the rate change, referred to as the final step:
    • The final step is set to 50kpps for all NIC to NIC tests and all L2 frame sizes except 9000B (changed from 100kpps used in previous releases).
    • The final step is set to 10kpps for all remaining tests, including 9000B and all vhost VM and memif Container tests.

All rates are reported from external Traffic Generator perspective.

2.1.6. Methodology: Packet Latency

TRex Traffic Generator (TG) is used for measuring latency of VPP DUTs. Reported latency values are measured using following methodology:

  • Latency tests are performed at 10%, 50% of discovered NDR rate (non drop rate) for each NDR throughput test and packet size (except IMIX).
  • TG sends dedicated latency streams, one per direction, each at the rate of 10kpps at the prescribed packet size; these are sent in addition to the main load streams.
  • TG reports min/avg/max latency values per stream direction, hence two sets of latency values are reported per test case; future release of TRex is expected to report latency percentiles.
  • Reported latency values are aggregate across two SUTs due to three node topology used for all performance tests; for per SUT latency, reported value should be divided by two.
  • 1usec is the measurement accuracy advertised by TRex TG for the setup used in FD.io labs used by CSIT project.
  • TRex setup introduces an always-on error of about 2*2usec per latency flow - additonal Tx/Rx interface latency induced by TRex SW writing and reading packet timestamps on CPU cores without HW acceleration on NICs closer to the interface line.

2.1.7. Methodology: KVM VM vhost

CSIT rls1801 introduced test environment configuration changes to KVM Qemu vhost-user tests in order to more representatively measure VPP-18.01 release performance in configurations with vhost-user interfaces and different Qemu settings.

FD.io CSIT performance lab is testing VPP vhost with KVM VMs using following environment settings:

  • Tests with varying Qemu virtio queue (a.k.a. vring) sizes: [vr256] default 256 descriptors, [vr1024] 1024 descriptors to optimize for packet throughput;
  • Tests with varying Linux CFS 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;
  • Resulting test cases are all combinations with [vr256,vr1024] and [cfs,cfsrr1] settings;
  • Adjusted Linux kernel CFS scheduler policy for data plane threads used in CSIT is documented in CSIT Performance Environment Tuning wiki. The purpose is to verify performance impact (NDR, PDR throughput) and same test measurements repeatability, by making VPP and VM data plane threads less susceptible to other Linux OS system tasks hijacking CPU cores running those data plane threads.

2.1.8. Methodology: LXC and Docker Containers memif

CSIT rls1801 introduced additional tests taking advantage of VPP memif virtual interface (shared memory interface) tests to interconnect VPP instances. VPP vswitch instance runs in bare-metal user-mode handling Intel x520 NIC 10GbE interfaces and connecting over memif (Master side) virtual interfaces to more instances of VPP running in LXC or in Docker Containers, both with memif virtual interfaces (Slave side). LXCs and Docker Containers run in a priviliged mode with VPP data plane worker threads pinned to dedicated physical CPU cores per usual CSIT practice. All VPP instances run the same version of software. This test topology is equivalent to existing tests with vhost-user and VMs as described earlier in Tested Physical Topologies.

More information about CSIT LXC and Docker Container setup and control is available in Container Orchestration in CSIT.

2.1.9. Methodology: Container Topologies Orchestrated by K8s

CSIT rls1801 introduced new tests of Container topologies connected over the memif virtual interface (shared memory interface). In order to provide simple topology coding flexibility and extensibility container orchestration is done with Kubernetes using Docker images for all container applications including VPP. Ligato is used to address the container networking orchestration that is integrated with K8s, including memif support.

For these tests VPP vswitch instance runs in a Docker Container handling Intel x520 NIC 10GbE interfaces and connecting over memif (Master side) virtual interfaces to more instances of VPP running in Docker Containers with memif virtual interfaces (Slave side). All Docker Containers run in a priviliged mode with VPP data plane worker threads pinned to dedicated physical CPU cores per usual CSIT practice. All VPP instances run the same version of software. This test topology is equivalent to existing tests with vhost-user and VMs as described earlier in Tested Physical Topologies.

More information about CSIT Container Topologies Orchestrated by K8s is available in Container Orchestration in CSIT.

2.1.10. Methodology: IPSec with Intel QAT HW cards

VPP IPSec performance tests are using DPDK cryptodev device driver in combination with HW cryptodev devices - Intel QAT 8950 50G - present in LF FD.io physical testbeds. DPDK cryptodev can be used for all IPSec data plane functions supported by VPP.

Currently CSIT rls1801 implements following IPSec test cases:

  • AES-GCM, CBC-SHA1 ciphers, in combination with IPv4 routed-forwarding with Intel xl710 NIC.
  • CBC-SHA1 ciphers, in combination with LISP-GPE overlay tunneling for IPv4-over-IPv4 with Intel xl710 NIC.

2.1.11. Methodology: TRex Traffic Generator Usage

TRex traffic generator is used for all CSIT performance tests. TRex stateless mode is used to measure NDR and PDR throughputs using binary search (NDR and PDR discovery tests) and for quick checks of DUT performance against the reference NDRs (NDR check tests) for specific configuration.

TRex is installed and run on the TG compute node. The typical procedure is:

  • If the TRex is not already installed on TG, it is installed in the suite setup phase - see TRex intallation.

  • TRex configuration is set in its configuration file

    /etc/trex_cfg.yaml

  • TRex is started in the background mode

    $ sh -c 'cd <t-rex-install-dir>/scripts/ && sudo nohup ./t-rex-64 -i -c 7 --iom 0 > /tmp/trex.log 2>&1 &' > /dev/null

  • There are traffic streams dynamically prepared for each test, based on traffic profiles. The traffic is sent and the statistics obtained using trex_stl_lib.api.STLClient.

Measuring packet loss

  • Create an instance of STLClient
  • Connect to the client
  • Add all streams
  • Clear statistics
  • Send the traffic for defined time
  • Get the statistics

If there is a warm-up phase required, the traffic is sent also before test and the statistics are ignored.

Measuring latency

If measurement of latency is requested, two more packet streams are created (one for each direction) with TRex flow_stats parameter set to STLFlowLatencyStats. In that case, returned statistics will also include min/avg/max latency values.

2.1.12. Methodology: TCP/IP tests with WRK tool

WRK HTTP benchmarking tool is used for experimental TCP/IP and HTTP tests of VPP TCP/IP stack and built-in static HTTP server. WRK has been chosen as it is capable of generating significant TCP/IP and HTTP loads by scaling number of threads across multi-core processors.

This in turn enables quite high scale benchmarking of the main TCP/IP and HTTP service including HTTP TCP/IP Connections-Per-Second (CPS), HTTP Requests-Per-Second and HTTP Bandwidth Throughput.

The initial tests are designed as follows:

  • HTTP and TCP/IP Connections-Per-Second (CPS)
    • WRK configured to use 8 threads across 8 cores, 1 thread per core.
    • Maximum of 50 concurrent connections across all WRK threads.
    • Timeout for server responses set to 5 seconds.
    • Test duration is 30 seconds.
    • Expected HTTP test sequence:
      • Single HTTP GET Request sent per open connection.
      • Connection close after valid HTTP reply.
      • Resulting flow sequence - 8 packets: >S,<S-A,>A,>Req,<Rep,>F,<F,> A.
  • HTTP Requests-Per-Second
    • WRK configured to use 8 threads across 8 cores, 1 thread per core.
    • Maximum of 50 concurrent connections across all WRK threads.
    • Timeout for server responses set to 5 seconds.
    • Test duration is 30 seconds.
    • Expected HTTP test sequence:
      • Multiple HTTP GET Requests sent in sequence per open connection.
      • Connection close after set test duration time.
      • Resulting flow sequence: >S,<S-A,>A,>Req[1],<Rep[1],..,>Req[n],<Rep[n],>F,<F,>A.