openstack
/
charm-deployment-guide
Code Issues Proposed changes

Update deploy/configure instructions for SR-IOV and HW-offload 

Related-Bug: #1908351
Related-Bug: #1907454
Change-Id: Ic9258bc4963a6e5ab23acc5c3980c969d74298c3
This commit is contained in:
Frode Nordahl 2021-03-16 16:59:53 +01:00
parent cbe50300e5
commit 2dd71d53d3
2 changed files with 18 additions and 5 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

11
deploy-guide/source/app-hardware-offload.rst
View File

@ -66,9 +66,7 @@ configured through MAAS.
.. note::
Use of VF-LAG halfs the offloaded port capacity of the card and as such VF
representor port configuration should only be made on the first network port
on the card.
Use of VF-LAG reduces the offloaded port capacity of the card by 50%.
Charm configuration
~~~~~~~~~~~~~~~~~~~
@ -97,7 +95,7 @@ The following overlay may be used with the OpenStack base deployment bundle:
charm: cs:neutron-openvswitch
options:
enable-hardware-offload: true
sriov-numvfs: "enp3s0f0:64 enp3s0f1:0"
sriov-numvfs: "enp3s0f0:32 enp3s0f1:32"
firewall-driver: openvswitch
neutron-api:
charm: cs:neutron-api
@ -117,6 +115,11 @@ The nova-compute charm is configured to use the VF functions provided by the
network cards using the ``pci-passthrough-whitelist`` option. The above example
demonstrates configuration for VXLAN overlay networking.
.. caution::
After deploying the above example the machines hosting neutron-openvswitch
units must be rebooted for the changes to take effect.
Creating hardware offloaded ports
~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~

12
deploy-guide/source/app-ovn.rst
View File

@ -140,7 +140,7 @@ deployment.
charm: cs:ovn-chassis
options:
enable-hardware-offload: true
sriov-numvfs: "enp3s0f0:64 enp3s0f1:0"
sriov-numvfs: "enp3s0f0:32 enp3s0f1:32"
neutron-api:
charm: cs:neutron-api
options:
@ -150,6 +150,11 @@ deployment.
options:
pci-passthrough-whitelist: '{"address": "*:03:*", "physical_network": null}'
.. caution::
After deploying the above example the machines hosting ovn-chassis
units must be rebooted for the changes to take effect.
Boot an instance
^^^^^^^^^^^^^^^^
@ -261,6 +266,11 @@ Enable SR-IOV, map physical network name 'physnet2' to the physical port named
juju config ovn-chassis sriov-device-mappings=physnet2:enp3s0f0
juju config ovn-chassis sriov-numvfs=enp3s0f0:4
.. caution::
After deploying the above example the machines hosting ovn-chassis
units must be rebooted for the changes to take effect.
After enabling the virtual functions you should take note of the ``vendor_id``
and ``product_id`` of the virtual functions: