Jump to: navigation, search


SR-IOV Networking in OpenStack Liberty

OpenStack Juno adds inbox support to request VM access to virtual network via SR-IOV NIC. With the introduction of SR-IOV based NICs, the traditional virtual bridge is no longer required. Each SR-IOV port is associated with a virtual function (VF). SR-IOV ports may be provided by Hardware-based Virtual Ethernet Bridging (HW VEB); or they may be extended to an upstream physical switch (IEEE 802.1br). There are two ways that SR-IOV port may be connected:

  • directly connected to its VF
  • connected with a macvtap device that resides on the host, which is then connected to the corresponding VF


Nova support for SR-IOV enables scheduling an instance with SR-IOV ports based on their network connectivity. The neutron ports' associated physical networks have to be considered in making the scheduling decision. PCI Whitelist has been enhanced to allow tags to be associated with PCI devices. PCI devices available for SR-IOV networking should be tagged with physical_network label.

For SR-IOV networking, a pre-defined tag "physical_network" is used to define the physical network to which the devices are attached. A whitelist entry is defined as:

   ["vendor_id": "<id>",] ["product_id": "<id>",]
   ["address": "[[[[<domain>]:]<bus>]:][<slot>][.[<function>]]" |
   "devname": "Ethernet Interface Name",]  
   "physical_network":"name string of the physical network"

<id> can be an asterisk (*) or a valid vendor/product ID as displayed by the Linux utility lspci. The address uses the same syntax as in lspci. The devname can be a valid PCI device name. The only device names that are supported are those displayed by the Linux utility ifconfig -a and correspond to either a PF or a VF on a vNIC.

If the device defined by the address or devname corresponds to a SR-IOV PF, all VFs under the PF will match the entry.

Multiple whitelist entries per host are supported.


Neutron support for SR-IOV requires ML2 Plugin with SR-IOV supporting mechanism driver. Currently there is ML2 Mechanism Driver for SR-IOV capable NIC based switching (HW VEB). There are network adapters from different vendors that vary by supporting various functionality. If VF link state update is supported by vendor network adapter, the SR-IOV NIC L2 agent should be deployed to leverage this functionality .

VM creation flow with SR-IOV vNIC

  • Create one or more neutron ports. Run:
  neutron port-create <net-id> --binding:vnic-type <direct | macvtap | normal>
  • Boot VM with one or more neutron ports. Run:
  nova boot --flavor m1.large --image <image>
         --nic port-id=<port1> --nic port-id=<port2> <vm name>

Note that in the nova boot API, users can specify either a port-ID or a net-ID. If a net-ID is specified, it is assumed that the user is requesting a normal virtual port (which is not an SR-IOV port).

SR-IOV Configuration

Controller Node

Nova Scheduler

Enabling PciPassthroughFilter modify /etc/nova/nova.conf

 scheduler_available_filters = nova.scheduler.filters.all_filters
 scheduler_default_filters = RetryFilter, AvailabilityZoneFilter, RamFilter, ComputeFilter, ComputeCapabilitiesFilter, ImagePropertiesFilter, PciPassthroughFilter

Neutron Server

Using ML2 Neutron plugin modify /etc/neutron/plugins/ml2/ml2_conf.ini:

tenant_network_types = vlan
type_drivers = vlan
mechanism_drivers = openvswitch,sriovnicswitch
network_vlan_ranges = physnet1:2:100

Add supported PCI vendor VF devices (Only needed on non Mellanox/Intel HCA's), defined by vendor_id:product_id according to the PCI ID Repository in the /etc/neutron/plugins/ml2/ml2_conf_sriov.ini:

# (ListOpt) Comma-separated list of
# supported Vendor PCI Devices, in format vendor_id:product_id
# supported_pci_vendor_devs = 15b3:1004, 8086:10ca
# Example: supported_pci_vendor_devs = 15b3:1004
supported_pci_vendor_devs = vendor_id:product_id

If SR-IOV network adapters support VF link state setting and admin state management is desired, make sure to add /etc/neutron/plugins/ml2/ml2_conf_sriov.ini [ml2_sriov] section the following setting(Deprecated in Liberty and from Mitaka will be enabled by default):

agent_required = True

Note:Usually the agent_required should be True unless you have Intel NIC which doesn't support admin state change.

edit neutron-server service (usually at: /usr/lib/systemd/system/neutron-server.service) Neutron server should be run with the configuration files /etc/neutron/plugins/ml2/ml2_conf_sriov.ini

neutron-server --config-file /usr/share/neutron/neutron-dist.conf --config-dir /usr/share/neutron/server --config-file /etc/neutron/neutron.conf --config-file /etc/neutron/plugin.ini --config-file /etc/neutron/plugins/ml2/ml2_conf_sriov.ini --config-dir /etc/neutron/conf.d/common --config-dir /etc/neutron/conf.d/neutron-server --log-file /var/log/neutron/server.log

Reload systemctl daemon

systemctl daemon-reload

Restart neutron-server

systemctl restart neutron-server.service

Compute Node

enable SR-IOV on network adapter

By default SRIOV is disabled, you will have to enable SRIOV and create the VFs on each compute host that should support SRIOV functionality. Currently specific Intel and Mellanox cards are known to support SRIOV. Below resources contain information on how to enable and create the VFs:


Nova-compute needs to know which PCI devices are allowed to be passed through to the VMs. Also for SRIOV PCI devices it needs to know to which physical network the VF belongs. This is done through the pci_passthrough_whitelist parameter under the default section in /etc/nova/nova.conf. For example if we want to whitelist and tag the VFs by their PCI address we would use the following setting:

pci_passthrough_whitelist = {"address":"*:0a:00.*","physical_network":"physnet1"}

This associates any VF with address that includes ':0a:00.' in its address to the physical network physnet1.

Alternatively we can also use the interface name and say: "All VFs created from eth2 are allowed to be passed through and belong to physical network physnet1" in this case eth2 is the PF. For example:

pci_passthrough_whitelist = {"devname": "eth2", "physical_network": "physnet1"}

After configuring the whitelist you have to restart nova-compute service.

systemctl restart openstack-nova-compute.service

When using devstack pci_passthrough_whitelist can be configured in local.conf file, for example:

pci_passthrough_whitelist = {"'"address"'":"'"*:02:00.*"'","'"physical_network"'":"'"default"'"}

SR-IOV neutron agent (Optional)

Only if the hardware supports it and you want to enable changing the port admin_state, you have to run the Neutron SR-IOV agent. Otherwise it's suggested to use agent_required = False.

Note:If you configured agent_required=True on the Neutron server, you must run the Agent on each compute node.

Install openstack-neutron-sriov-nic-agent.noarch

yum install -y openstack-neutron-sriov-nic-agent.noarch

Modify "physical_device_mappings" at /etc/neutron/plugins/ml2/sriov_agent.ini: Where:

  • physnet1 is the physical network
  • eth1 is the physical function (PF)
  • exclude_devices is empty so all the VFs associated with eth1 may be configured by the agent
# (ListOpt) Comma-separated list of <physical_network>:<network_device>
# tuples mapping physical network names to the agent's node-specific
# physical network device interfaces of SR-IOV physical function to be used
# for VLAN networks. All physical networks listed in network_vlan_ranges on
# the server should have mappings to appropriate interfaces on each agent.
# physical_device_mappings =
# Example: physical_device_mappings = physnet1:eth1
physical_device_mappings = default:p2p1

Add/modify the following lines at the end of the file.

firewall_driver = neutron.agent.firewall.NoopFirewallDriver

Note:SR-IOV agent only work with NoopFirewallDriver when Security Groups are enabled, but you can still use other firewall_driver for other Agents by updating their conf with the requested firewall driver.

Enable Neutron SR-IOV agent. Run:

systemctl enable neutron-sriov-nic-agent.service

Start Neutron SR-IOV agent. Run:

systemctl start neutron-sriov-nic-agent.service

Exclude VFs

If you want to exclude some of the VFs so the agent does not configure them, you need to list them in the sriov_nic section:

Example: exclude_devices = eth1:0000:07:00.2; 0000:07:00.3, eth2:0000:05:00.1; 0000:05:00.2