Jump to: navigation, search

Difference between revisions of "Mellanox-Neutron-Train-Redhat-InfiniBand"

(Partition membership configuration)
(Blanked the page)
 
(8 intermediate revisions by the same user not shown)
Line 1: Line 1:
=Overview=
 
  
==Mellanox Neutron ML2 Driver==
 
 
Mellanox ML2 Mechanism Driver implements the ML2 Plugin Mechanism Driver API.
 
This driver supports Mellanox embedded switch functionality as part of the InfiniBand HCA. Mellanox ML2 Mechanism Driver provides functional parity with Mellanox Neutron plugin.
 
 
Mellanox ML2 Mechanism Driver supports DIRECT (pci passthrough) vnic type to allow VM network connectivity.
 
For vnic type configuration API details, please refer to configuration reference guide.
 
Hardware vNICs mapped to the guest VMs allow higher performance and advanced features such as RDMA (remote direct memory access).
 
 
The driver supports VLAN network type to facilitate virtual networks either on InfiniBand fabrics.
 
 
===Prerequisites===
 
* CentOS 7.6 / Ubuntu 18.04 or later
 
* Mellanox ConnectX® Family device:
 
    [https://www.mellanox.com/page/products_dyn?product_family=119  ConnectX®-3/ConnectX®-3 PRO]
 
    [https://www.mellanox.com/page/products_dyn?product_family=201& ConnectX®-4]/[https://www.mellanox.com/page/products_dyn?product_family=214&mtag=connectx_4_lx_en_ic ConnectX®-4Lx]
 
    [https://www.mellanox.com/page/products_dyn?product_family=258&mtag=connectx_5_vpi_card ConnectX®-5]
 
    [https://www.mellanox.com/page/products_dyn?product_family=265&mtag=connectx_6_vpi_card ConnectX®-6]
 
* Driver: [http://www.mellanox.com/page/products_dyn?product_family=26&mtag=linux_sw_drivers Mellanox OFED] 4.6-1.0.1.1 or greater
 
* A running OpenStack environment installed (RDO Manager or Packstack).
 
* [https://community.mellanox.com/docs/DOC-1317 SR-IOV] enabled on all compute nodes.
 
* The software package [http://www.linuxfoundation.org/collaborate/workgroups/networking/iproute2 iproute2] installed on all Compute nodes
 
* [http://www.mellanox.com/page/products_dyn?product_family=100 Mellanox UFM] greater than 5.9.5
 
 
=InfiniBand Network=
 
An Infiniband network relies on a software entity to manage the network.
 
This entity is referred to as a Subnet Manager or SM.
 
Mellanox Neutron Plugin uses InfiniBand Partitions (PKeys) to separate Networks.
 
 
==SM Node==
 
 
=== OpenSM Provisioning with SDN Mechanism Driver ===
 
 
[https://community.mellanox.com/docs/DOC-2155 SDN Mechanism Driver] allows OpenSM dynamically assign PKs in the IB network.
 
 
More details about applying SDN Mechanism Driver with [http://www.mellanox.com/page/products_dyn?product_family=220&mtag=mellanox_neo NEO] can be found [https://community.mellanox.com/docs/DOC-2251 here]
 
 
=== Manual OpenSM Configuration ===
 
 
====For ConnectX®-3/ConnectX®-3Pro use the following configuration====
 
 
Change the following in /etc/opensm/opensm.conf:
 
  allow_both_pkeys TRUE
 
 
====For ConnectX®-4 or newer use the following configuration====
 
 
Change the following in /etc/opensm/opensm.conf:
 
virt_enabled 2
 
 
==== Partition membership configuration ====
 
In an Infiniband network, network segmentation is achieved through partitions (roughly equivalent to ETH VLAN). Each partition has its own key (15bit partition key).
 
This file contains configuration of partition membership for all network endpoints in the system.
 
 
The following guidelines need to be followed to allow connectivity
 
 
For each PKEY its required to:
 
1. Have OpenSM PF(physical function IB device) GUID be a member of the partition.
 
2. Have L3 and DHCP PF GUID be a member of the partition.
 
3. Have VF GUID be member of the partition (of the relevant VMs).
 
 
There is a 1:1 mapping between VLAN assigned to a network in openstack and the translation of that VLAN to PKEY (e.g VLAN 10 translates to PKEY 10)
 
 
===== Example1: =====
 
 
The following configurations corresponds to a single vlan network (vlan 3) with two VMs and a DHCP agent
 
 
management=0x7fff,ipoib, defmember=full : ALL, ALL_SWITCHES=full,SELF=full;
 
vlan3_vm1=0x3, indx0, ipoib, defmember=full : 0xfa163e0000c0851b;
 
vlan3_vm2=0x3, indx0, ipoib, defmember=full : 0xfa163e0000df0519;
 
vlan3_l3_dhcp_services=0x3, ipoib, defmember=full : 0xe41d2d030061f5fa;
 
vlan3_sm=0x3, ipoib, defmember=full : SELF;
 
 
Going over it line by line:
 
 
#1 make all member of the default (management) pkey
 
#2 VM1 port guid (VF) member of network with pkey 3
 
#3 VM2 port guid (VF) member of network with pkey 3
 
#4 L3/DHCP agent PF guid member of network with pkey3
 
#5 Subnet manager member of network with pkey3
 
 
===== Example 2: =====
 
 
The following configurations corresponds to a two vlan networks (vlan 3, vlan 5) with one VM on each network and a DHCP and L3 agent
 
 
management=0x7fff,ipoib, defmember=full : ALL, ALL_SWITCHES=full,SELF=full;
 
vlan3_vm1=0x3, indx0, ipoib, defmember=full : 0xfa163e0000c0851b;
 
vlan5_vm2=0x5, indx0, ipoib, defmember=full : 0xfa163e0000df0519;
 
vlan3_l3_dhcp_services=0x3, ipoib, defmember=full : 0xe41d2d030061f5fa;
 
vlan5_l3_dhcp_services=0x5, ipoib, defmember=full : 0xe41d2d030061f5fa;
 
vlan3_sm=0x3, ipoib, defmember=full : SELF;
 
vlan5_sm=0x5, ipoib, defmember=full : SELF;
 
 
Going over it line by line:
 
 
#1 make all member of the default (management) pkey
 
#2 VM1 port guid (VF) member of network with pkey 3
 
#3 VM2 port guid (VF) member of network with pkey 5
 
#4 L3/DHCP agent PF guid member of network with pkey3
 
#5 L3/DHCP agent PF guid member of network with pkey5
 
#6 Subnet manager member of network with pkey3
 
#7 Subnet manager member of network with pkey5
 
 
4. Restart the OpenSM:
 
# systemctl restart opensmd.service
 
 
==Controller Node==
 
 
To configure the Controller node:
 
 
1. Install Mellanox RPMs:
 
# yum install -y --nogpgcheck python-networking-mlnx
 
 
===Neutron Server===
 
1. Make sure ML2 is the current Neutron plugin by checking the core_plugin parameter in '''/etc/neutron/neutron.conf''':
 
core_plugin = neutron.plugins.ml2.plugin.Ml2Plugin
 
 
2. Make sure '''/etc/neutron/plugin.ini''' is pointing to '''/etc/neutron/plugins/ml2/ml2_conf.ini''' (symbolic link)
 
 
3. Modify '''/etc/neutron/plugins/ml2/ml2_conf.ini''' by adding the following:
 
[ml2]
 
type_drivers = vlan,flat
 
tenant_network_types = vlan
 
mechanism_drivers = mlnx_infiniband, openvswitch
 
# or mechanism_drivers = mlnx_infiniband, linuxbridge
 
[ml2_type_vlan]
 
network_vlan_ranges = default:1:10
 
 
4. Start (or restart) the Neutron server:
 
# systemctl restart neutron-server.service
 
 
===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
 
 
== Network Node==
 
To configure the Network node:
 
 
Prerequisites:
 
E_IPoIB port is configured and up
 
 
1. Make sure that eIPoIB module is up and configured in '''/etc/infiniband/openib.conf''': For more information, please refer to eIPoIB configuration in Mellanox OFED User Manual.
 
E_IPOIB_LOAD=yes
 
 
 
2. Restart openibd:
 
# service openibd restart
 
 
3. Modify the network bridge configuration according to the use of OpenVswitch or LinuxBridge
 
* 3.1 OpenVswitch '''/etc/neutron/plugins/ml2/openvswitch_agent.ini'''
 
[ovs]
 
bridge_mappings = default:br-<eIPoIB interface>
 
* 3.2 LinuxBridge file '''/etc/neutron/plugins/ml2/linuxbridge_agent.ini''':
 
[linux_bridge]
 
physical_interface_mappings = default:<eIPoIB interface>
 
 
 
NOTE: In order to obtain the eIPoIB interface name, run the ethtool tool command (see below) and check that driver name is eth_ipoib:
 
# ethtool -i <eIPoIB_interface>
 
driver: eth_ipoib
 
.....
 
 
 
4. Restart network bridge:
 
* 4.1 OpenVswitch
 
systemctl restart neutron-openvswitch-agent.service
 
* 4.2 LinuxBridge (if using linux bridge)
 
# service neutron-linuxbridge-agent restart
 
 
 
NOTE: For DHCP support, the Network node should use the Mellanox Dnsmasq driver as the DHCP driver.
 
 
===DHCP Server (Usually part of the Network node)===
 
 
1. Modify '''/etc/neutron/dhcp_agent.ini''' as follows and according to OVS or Linuxbridge:
 
dhcp_driver = networking_mlnx.dhcp.mlnx_dhcp.MlnxDnsmasq
 
dhcp_broadcast_reply = True
 
 
1.1 For OVS
 
    interface_driver =  neutron.agent.linux.interface.OVSInterfaceDriver
 
1.2 For Linux Bridge
 
      interface_driver =  neutron.agent.linux.interface.BridgeInterfaceDriver
 
 
 
2. Restart DHCP server:
 
# systemctl restart neutron-dhcp-agent.service
 
 
==Compute Nodes==
 
 
To configure the Compute Node:
 
 
1. Install Mellanox RPMs:
 
# yum install --nogpgcheck -y python-networking-mlnx
 
 
2. Create the file '''/etc/modprobe.d/mlx4_ib.conf''' and add the following: 
 
options mlx4_ib sm_guid_assign=0
 
 
3. Restart the driver:
 
# /etc/init.d/openibd restart
 
 
===Nova Compute===
 
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":"default"}
 
This associates any VF with address that includes ':0a:00.' in its address to the physical network default.
 
 
1. add  pci passthrough_whitelist to '''/etc/nova/nova.conf'''
 
 
2. Restart Nova:
 
# systemctl restart openstack-nova-compute
 
 
===Neutron MLNX Agent===
 
1. Run:
 
# systemctl enable neutron-mlnx-agent.service
 
# systemctl start neutron-mlnx-agent.service
 
 
2. Run:
 
# systemctl daemon-reload
 
 
3. In the file '''/etc/neutron/plugins/mlnx/mlnx_conf.ini''', the parameters tenant_network_type , and network_vlan_ranges should be configured as the controllers:
 
physical_interface_mappings = default:<ib_interface>(for example default:ib0)
 
 
4. Modify the file '''/etc/neutron/plugins/ml2/eswitchd.conf''' as follows:
 
fabrics = default:<ib_interface> (for example default:ib0)
 
 
5. Start eSwitchd:
 
# systemctl enable eswitchd.service
 
# systemctl start  eswitchd.service
 
 
6. Start the Neutron agent:
 
# systemctl restart neutron-mlnx-agent
 
 
==Known issues and Troubleshooting==
 
For known issues and troubleshooting options refer to [https://community.mellanox.com/docs/DOC-1127 Mellanox OpenStack Troubleshooting]
 
 
Issue: Missing zmq package on all nodes (Controller/Compute)
 
Solution:
 
<pre>
 
# wget https://bootstrap.pypa.io/get-pip.py
 
# sudo python get-pip.py
 
# sudo pip install pyzmq
 
</pre>
 

Latest revision as of 12:04, 5 September 2019