208. Eventdev Pipeline Perf Tests

208.1. Prerequisites

Each of the 10Gb/25Gb/40Gb/100Gb Ethernet* ports of the DUT is directly connected in full-duplex to a different port of the peer Ixia ports(traffic generator).

Using TCL commands, the Ixia can be configured to send and receive traffic on a given set of ports.

If using vfio the kernel must be >= 3.6+ and VT-d must be enabled in bios.When using vfio, use the following commands to load the vfio driver and bind it to the device under test

modprobe vfio
modprobe vfio-pci
usertools/dpdk-devbind.py --bind=vfio-pci device_bus_id
usertools/dpdk-devbind.py --bind=vfio-pci eventdev_device_bus_id

208.2. Create huge pages

mkdir -p /dev/huge mount -t hugetlbfs none /dev/huge echo 24 > /proc/sys/vm/nr_hugepages

208.3. Configure limits of Eventdev devices

Set all eventdev devices sso and ssow limits to zero. Then set eventdev device under tests sso and ssow limits to non-zero values as per cores/queues requriments

echo 0 > /sys/bus/pci/devices/eventdev_device_bus_id/limits/sso
echo 0 > /sys/bus/pci/devices/eventdev_device_bus_id/limits/ssow

Example

echo 0 > /sys/bus/pci/devices/eventdev_device_bus_id/limits/tim
echo 1 > /sys/bus/pci/devices/eventdev_device_bus_id/limits/npa
echo 16 > /sys/bus/pci/devices/eventdev_device_bus_id/limits/sso
echo 32 > /sys/bus/pci/devices/eventdev_device_bus_id/limits/ssow
  • eventdev_device_bus_id/limits/sso : Max limit 256
  • eventdev_device_bus_id/limits/ssow: Max limit 52

208.4. Test Case: Performance 1port atomic test

Description: Execute performance test with Atomic_atq type of stage in multi-flow situation for various cores.

  1. Run the sample with below command:

    # ./build/dpdk-eventdev_pipeline -c 0xe00000 -w eventdev_device_bus_id -w device_bus_id -- -w 0xc00000 -n=0 --dump
    

    Parameters:

    -c, COREMASK         : Hexadecimal bitmask of cores to run on
    -w, --pci-whitelist  : Add a PCI device in white list.
                           Only use the specified PCI devices. The argument format
                           is <[domain:]bus:devid.func>. This option can be present
                           several times (once per device).
    EAL Commands
    -w, --worker-mask=core mask : Run worker on CPUs in core mask
    -n, --packets=N             : Send N packets (default ~32M), 0 implies no limit
    -D, --dump                   Print detailed statistics before exit
    
  2. Use Ixia to send huge number of packets(with same 5-tuple and different 5-tuple)

  3. Observe the speed of packets received(Rx-rate) on Ixia.

208.5. Test Case: Performance 1port parallel test

Description: Execute performance test with Parallel_atq type of stage in multi-flow situation for various cores.

  1. Run the sample with below command:

    # ./build/dpdk-eventdev_pipeline -c 0xe00000 -w eventdev_device_bus_id -w device_bus_id -- -w 0xc00000 -n=0 -p --dump
    

    Parameters:

    -c, COREMASK         : Hexadecimal bitmask of cores to run on
    -w, --pci-whitelist  : Add a PCI device in white list.
                           Only use the specified PCI devices. The argument format
                           is <[domain:]bus:devid.func>. This option can be present
                           several times (once per device).
    EAL Commands
    -w, --worker-mask=core mask : Run worker on CPUs in core mask
    -n, --packets=N             : Send N packets (default ~32M), 0 implies no limit
    -p, --parallel              : Use parallel scheduling
    -D, --dump                  : Print detailed statistics before exit
    
  2. Use Ixia to send huge number of packets(with same 5-tuple and different 5-tuple)

  3. Observe the speed of packets received(Rx-rate) on Ixia.

208.6. Test Case: Performance 1port ordered test

Description: Execute performance test with Ordered_atq type of stage in multi-flow situation for various cores.

  1. Run the sample with below command:

    # ./build/dpdk-eventdev_pipeline -c 0xe00000 -w eventdev_device_bus_id -w device_bus_id -- -w 0xc00000 -n=0 -o --dump
    

    Parameters:

    -c, COREMASK         : Hexadecimal bitmask of cores to run on
    -w, --pci-whitelist  : Add a PCI device in white list.
                           Only use the specified PCI devices. The argument format
                           is <[domain:]bus:devid.func>. This option can be present
                           several times (once per device).
    EAL Commands
    -w, --worker-mask=core mask : Run worker on CPUs in core mask
    -n, --packets=N             : Send N packets (default ~32M), 0 implies no limit
    -o, --ordered                Use ordered scheduling
    -D, --dump                  : Print detailed statistics before exit
    
  2. Use Ixia to send huge number of packets(with same 5-tuple and different 5-tuple)

  3. Observe the speed of packets received(Rx-rate) on Ixia.

208.7. Test Case: Performance 2port atomic test

Description: Execute performance test with Atomic_atq type of stage in multi-flow situation for various cores.

  1. Run the sample with below command:

    # ./build/dpdk-eventdev_pipeline -c 0xe00000 -w eventdev_device_bus_id -w device0_bus_id -w device1_bus_id -- -w 0xc00000 -n=0 --dump
    

    Parameters:

    -c, COREMASK         : Hexadecimal bitmask of cores to run on
    -w, --pci-whitelist  : Add a PCI device in white list.
                           Only use the specified PCI devices. The argument format
                           is <[domain:]bus:devid.func>. This option can be present
                           several times (once per device).
    EAL Commands
    -w, --worker-mask=core mask : Run worker on CPUs in core mask
    -n, --packets=N             : Send N packets (default ~32M), 0 implies no limit
    -D, --dump                   Print detailed statistics before exit
    
  2. Use Ixia to send huge number of packets(with same 5-tuple and different 5-tuple)

  3. Observe the speed of packets received(Rx-rate) on Ixia.

208.8. Test Case: Performance 2port parallel test

Description: Execute performance test with Parallel_atq type of stage in multi-flow situation for various cores.

  1. Run the sample with below command:

    # ./build/dpdk-eventdev_pipeline -c 0xe00000 -w eventdev_device_bus_id -w device0_bus_id -w device1_bus_id -- -w 0xc00000 -n=0 -p --dump
    

    Parameters:

    -c, COREMASK         : Hexadecimal bitmask of cores to run on
    -w, --pci-whitelist  : Add a PCI device in white list.
                           Only use the specified PCI devices. The argument format
                           is <[domain:]bus:devid.func>. This option can be present
                           several times (once per device).
    EAL Commands
    -w, --worker-mask=core mask : Run worker on CPUs in core mask
    -n, --packets=N             : Send N packets (default ~32M), 0 implies no limit
    -p, --parallel              : Use parallel scheduling
    -D, --dump                  : Print detailed statistics before exit
    
  2. Use Ixia to send huge number of packets(with same 5-tuple and different 5-tuple)

  3. Observe the speed of packets received(Rx-rate) on Ixia.

208.9. Test Case: Performance 2port ordered test

Description: Execute performance test with Ordered_atq type of stage in multi-flow situation for various cores.

  1. Run the sample with below command:

    # ./build/dpdk-eventdev_pipeline -c 0xe00000 -w eventdev_device_bus_id -w device0_bus_id -w device1_bus_id -- -w 0xc00000 -n=0 -o --dump
    

    Parameters:

    -c, COREMASK         : Hexadecimal bitmask of cores to run on
    -w, --pci-whitelist  : Add a PCI device in white list.
                           Only use the specified PCI devices. The argument format
                           is <[domain:]bus:devid.func>. This option can be present
                           several times (once per device).
    EAL Commands
    -w, --worker-mask=core mask : Run worker on CPUs in core mask
    -n, --packets=N             : Send N packets (default ~32M), 0 implies no limit
    -o, --ordered                Use ordered scheduling
    -D, --dump                  : Print detailed statistics before exit
    
  2. Use Ixia to send huge number of packets(with same 5-tuple and different 5-tuple)

  3. Observe the speed of packets received(Rx-rate) on Ixia.

208.10. Test Case: Performance 4port atomic test

Description: Execute performance test with Atomic_atq type of stage in multi-flow situation for various cores.

  1. Run the sample with below command:

    # ./build/dpdk-eventdev_pipeline -c 0xe00000 -w eventdev_device_bus_id -w device0_bus_id -w device1_bus_id -w device2_bus_id -w device3_bus_id -- -w 0xc00000 -n=0 --dump
    

    Parameters:

    -c, COREMASK         : Hexadecimal bitmask of cores to run on
    -w, --pci-whitelist  : Add a PCI device in white list.
                           Only use the specified PCI devices. The argument format
                           is <[domain:]bus:devid.func>. This option can be present
                           several times (once per device).
    EAL Commands
    -w, --worker-mask=core mask : Run worker on CPUs in core mask
    -n, --packets=N             : Send N packets (default ~32M), 0 implies no limit
    -D, --dump                   Print detailed statistics before exit
    
  2. Use Ixia to send huge number of packets(with same 5-tuple and different 5-tuple)

  3. Observe the speed of packets received(Rx-rate) on Ixia.

208.11. Test Case: Performance 4port parallel test

Description: Execute performance test with Parallel_atq type of stage in multi-flow situation for various cores.

  1. Run the sample with below command:

    # ./build/dpdk-eventdev_pipeline -c 0xe00000 -w eventdev_device_bus_id -w device0_bus_id -w device1_bus_id -w device2_bus_id -w device3_bus_id -- -w 0xc00000 -n=0 -p --dump
    

    Parameters:

    -c, COREMASK         : Hexadecimal bitmask of cores to run on
    -w, --pci-whitelist  : Add a PCI device in white list.
                           Only use the specified PCI devices. The argument format
                           is <[domain:]bus:devid.func>. This option can be present
                           several times (once per device).
    EAL Commands
    -w, --worker-mask=core mask : Run worker on CPUs in core mask
    -n, --packets=N             : Send N packets (default ~32M), 0 implies no limit
    -p, --parallel              : Use parallel scheduling
    -D, --dump                  : Print detailed statistics before exit
    
  2. Use Ixia to send huge number of packets(with same 5-tuple and different 5-tuple)

  3. Observe the speed of packets received(Rx-rate) on Ixia.

208.12. Test Case: Performance 4port ordered test

Description: Execute performance test with Ordered_atq type of stage in multi-flow situation for various cores.

  1. Run the sample with below command:

    # ./build/dpdk-eventdev_pipeline -c 0xe00000 -w eventdev_device_bus_id -w device0_bus_id -w device1_bus_id -w device2_bus_id -w device3_bus_id -- -w 0xc00000 -n=0 -o --dump
    

    Parameters:

    -c, COREMASK         : Hexadecimal bitmask of cores to run on
    -w, --pci-whitelist  : Add a PCI device in white list.
                           Only use the specified PCI devices. The argument format
                           is <[domain:]bus:devid.func>. This option can be present
                           several times (once per device).
    EAL Commands
    -w, --worker-mask=core mask : Run worker on CPUs in core mask
    -n, --packets=N             : Send N packets (default ~32M), 0 implies no limit
    -o, --ordered                Use ordered scheduling
    -D, --dump                  : Print detailed statistics before exit
    
  2. Use Ixia to send huge number of packets(with same 5-tuple and different 5-tuple)

  3. Observe the speed of packets received(Rx-rate) on Ixia.