I'm just asking this question as a way to collate some learnings from support to users who have done this.
That's a great question, Billy ;)
Firstly, use OPNFV Fuel Brahamputra SR3 .iso from https://www.opnfv.org/opnfv-brahmaput...
After the .iso installation Use version 0.0.3 of the OVSNFV plugin (version 0.0.1 has a bug whereby VMs cannot be assigned public ip addresses, there is no version 0.0.2)
http://artifacts.opnfv.org/ovsnfv/bra... sha1sum: 7f719fad028bded41be0c6579ba3246010258942
The plugin has to be downloaded and installed separately using
fuel-master# fuel plugins --install opnfv-fuel-plugin-ovsnfv-brahmaputra.3.0.rpm
There are no bios settings that require to be changed for ovsnfv plugin. The installation will take care of hugetables, configuring neutron and so on.
The compute nodes will require at least 3 NICs – one for PXE, one for the DPDK poll mode driver (even though no traffic will be routed over this NIC – this is a restriction that will be removed in Colorado) and one for the remaining networks (storage, public, etc).
Use the repos on the fuel-master as per http://artifacts.opnfv.org/fuel/brahm...
Choose the name of an interface that has no other logical networks assigned to it. This is the name that is inserted into the 'NIC for DPDK' configuration for the plugin under Settings > Other > Userspace OVS support.
Set the hypervisor type to KVM and use baremetal hosts.
Known issues & Limtiations:
Fuel still configures so that all traffic from the host will go through kernel controlled interfaces. However ovs-dpdk still requires an interface in order to run it’s poll mode driver against. Therefore the compute host needs a spare/dummy interface that can be allocated to ovs-dpdk – this is specified on the plugin configuration page in the Settings tab under the Fuel UI before deployment. C release of OPNFV will also route tenant traffic from the physical n/w via the PMD.
0.0.3 OVSNFV still accelerates traffic to the guest so tenant traffic from the physical n/w still benefits somewhat and traffic between VMs on the same node benefits a lot.
Only Neutron with VLAN segmented tenant networks is supported.
The plugin has not been tested on virtual hosts.
An issue was reported on a single socket system so if you’re compute hosts are not NUMA you will need to edit the openvswitch service /etc/init.d/openvswitch-switch and modify '512,512' to just '512'.
Testing the installation:
Generally I test by creating a few VMs and pinging between them and or pinging them on their public ip addresses.
The built in self-tests in Fuel should pass except for one test regarding suspension/resumption of stacks. And some other tests that are not relevant - HA tests and security tests (default passwords in use).
Q. Do I need to configure huge pages on the compute node or change the kernel parameters on the environment settings?
A. No. Do not add any hugepage configuration to the kernel parameters for the environment in Fuel or manually modify the kernel command line on any of the hosts. The user-space enabled ... (more)
Asked: 2016-06-17 06:03:58 -0700
Seen: 896 times
Last updated: Jun 17 '16
© 2014 OPNFV. All Rights Reserved.
Linux Foundation is a registered trademark of The Linux Foundation.
Linux is a registered trademark of Linus Torvalds.