VXLAN support for OVS-DPDK

https://blueprints.launchpad.net/fuel/+spec/vxlan-support-for-ovs-dpdk

We want to utilize a VXLAN-based networking with OVS/DPDK for a high-performance scalable tenant networking.

Problem description

Currently, OVS-DPDK supports only VLAN segmentation. With VXLAN-based network segmentation being adopted more widely, supporting VXLAN with OVS-DPDK is very important for all NFV use cases.

Proposed changes

Currently, when using the VLAN-based networking with OVS-DPDK, we use the br-prv bridge but we do not assign an IP to it.

To implement a VXLAN-based segmentation with DPDK, we should use a br-mesh bridge whose configuration corresponds to the one of the br-prv bridge in case of VLAN with DPDK.

Web UI

None

Nailgun

The following changes are required in Nailgun:

• Remove restrictions for using DPDK in VXLAN-based segmentation case.
• Fix the Nailgun network serializer to generate transformations as described in the Data model section.

Data model

When operator enables DPDK for a particular interface with VXLAN-based segmentation to use it for the Private network, astute.yaml will be extended as follows:

• The network transformations field should include a vendor-specific attribute datapath_type: netdev for the br-mesh bridge:

  network_scheme:
    transformations:
    - action: add-br
      name: br-mesh
      provider: ovs
      vendor_specific:
        vlan_id: netgroup['vlan']
        datapath_type: netdev
  

• An interface should be added directly into the OVS br-mesh bridge using the add-port action with provider: dpdkovs:

  network_scheme:
    transformations:
    - action: add-port
      name: enp1s0f0
      bridge: br-mesh
      provider: dpdkovs
  

  No VLAN tag can be used here.

• A bond should be added directly into the OVS br-mesh bridge using the add-bond action with provider: dpdkovs:

  network_scheme:
    transformations:
    - action: add-bond
      bridge: br-mesh
      provider: dpdkovs
      bond_properties:
        mode: balance-rr
      interfaces:
      - enp1s0f0
      - enp1s0f1
      name: bond0
  

  No VLAN tag can be used here.

REST API

None

Orchestration

None

RPC Protocol

None

Fuel Client

None

Plugins

None

Fuel Library

To achieve VLAN-tagged VXLAN, the vendor specific attribute vlan_id for add-br should be converted to ovs-vsctl set port br-mesh tag=<vlan_id>.

Alternatives

Continue using the VLAN-based network segmentation.

Upgrade impact

None

Security impact

None

Notifications impact

None

End user impact

None

Performance impact

Performance impact is not expected.

Deployment impact

This feature requires using the VXLAN segmentation and a dedicated DPDK-capable network interface for the Private network.

Developer impact

None

Infrastructure impact

• The feature will be tested on a virtual environment.
• The performance testing will be conducted on a hardware environment

Documentation impact

The user guide should be updated according to the described feature.

Implementation

Assignee(s)

Primary assignee:

Anastasia Balobashina <atolochkova@mirantis.com>

Mandatory design review:

Aleksey Kasatkin <akasatkin@mirantis.com> Sergey Matov <smatov@mirantis.com>

Work Items

• Remove restrictions for using DPDK in VXLAN-based segmentation case.
• Fix the network serializer so that the transformations are configured as described in the Data model section.
• Convert the vendor specific attribute vlan_id for add-br to ovs-vsctl set port br-mesh tag=<vlan_id>.
• Test manually.
• Create a system test for DPDK.
• Verify the acceptance criteria.

Dependencies

None

Testing, QA

• API/CLI test cases for configuring the DPDK with VXLAN segmentation.
• Web UI test cases for configuring the DPDK with VXLAN segmentation.
• Test case for DPDK with VXLAN segmentation being discovered and configured properly.
• Test case for using the multiple-node network groups.
• Functional testing.
• Performance testing.

Acceptance criteria

• Ability to run a DPDK application on top of OVS/DPDK + VXLAN-enabled host
• A 3 Mpps packet rate on the 64-bytes UDP traffic on a single PMD thread multiplied by a number of DPDK cores.
• Ability to work on the 40 Gb and 2x10 cards from Intel’s Forteville family.

References

None