Open vSwitch Networks

This guide describes how to use the Open vSwitch network drivers. They provide network isolation using VLANs by tagging ports and basic network filtering using OpenFlow. Other traffic attributes that may be configured through Open vSwitch are not modified.

The VLAN ID will be the same for every interface in a given network, calculated automatically by OpenNebula. It may also be forced by specifying an VLAN_ID parameter in the Virtual Network template.

OpenNebula Configuration

The VLAN ID is calculated according to this configuration option /etc/one/oned.conf:

#  VLAN_IDS: VLAN ID pool for the automatic VLAN_ID assigment. This pool
#  is for 802.1Q networks (Open vSwitch and 802.1Q drivers). The driver
#  will try first to allocate VLAN_IDS[START] + VNET_ID
#     start: First VLAN_ID to use
#     reserved: Comma separated list of VLAN_IDs or ranges. Two numbers
#     separated by a colon indicate a range.

VLAN_IDS = [
    START    = "2",
    RESERVED = "0, 1, 4095"
]

By modifying this section, you can reserve some VLANs so they aren’t assigned to a Virtual Network. You can also define the first VLAN ID. When a new isolated network is created, OpenNebula will find a free VLAN ID from the VLAN pool. This pool is global and it’s also shared with the 802.1Q Networks.

The following configuration parameters can be adjusted in /var/lib/one/remotes/etc/vnm/OpenNebulaNetwork.conf:

Defining Open vSwitch Network

To create an Open vSwitch network, include the following information:

For example, you can define an Open vSwitch Network with the following template:

NAME    = "private4"
VN_MAD  = "ovswitch"
BRIDGE  = vbr1
VLAN_ID = 50          # Optional
...

Multiple VLANs (VLAN trunking)

VLAN trunking is also supported by adding the following tag to the NIC element in the VM template or to the Virtual Network template:

• VLAN_TAGGED_ID: Specify a range of VLANs to tag, for example: 1,10,30,32,100-200.

Using Open vSwitch on VXLAN Networks

This section describes how to use Open vSwitch on VXLAN networks. To use VXLAN you need to use a specialized version of the Open vSwitch driver that incorporates the features of the VXLAN driver. It’s necessary to be familiar with these two drivers, their configuration options, benefits, and drawbacks.

The VXLAN overlay network is used as a base with the Open vSwitch (instead of regular Linux bridge) on top. Traffic on the lowest level is isolated by the VXLAN encapsulation protocol and Open vSwitch still allows second level isolation by 802.1Q VLAN tags inside the encapsulated traffic. The main isolation is always provided by VXLAN, not 802.1Q VLANs. If 802.1Q is required to isolate the VXLAN, the driver needs to be configured with a user-created 802.1Q-tagged physical interface.

This hierarchy is important.

OpenNebula Configuration

There is no configuration specific to this driver except the options specified above and in the VXLAN Networks guide.

Defining an Open vSwitch - VXLAN Network

To create a network, include the following information:

For example, you can define an Open vSwitch - VXLAN Network with the following template:

NAME          = "private5"
VN_MAD        = "ovswitch_vxlan"
PHYDEV        = eth0
BRIDGE        = ovsvxbr0.10000
OUTER_VLAN_ID = 10000               # VXLAN VNI
VLAN_ID        = 50                 # Optional VLAN ID
...

In this example, the driver will check for the existence of bridge ovsvxbr0.10000. If it doesn’t exist, it will be created. Also, the VXLAN interface eth0.10000 will be created and attached to the Open vSwitch bridge ovsvxbr0.10000. When a Virtual Machine is instantiated, its bridge ports will be tagged with 802.1Q VLAN ID 50.

Open vSwitch with DPDK

This section describes how to use a DPDK datapath with the Open vSwitch drivers. When using the DPDK backend, the OpenNebula drivers will automatically configure the bridges and ports accordingly.

Please consider the following when using the DPDK datapath for Open vSwitch:

• An Open vSwitch version compiled with DPDK support is required.
• The VMs need to use the virtio interface for their NICs.
• Hugepages needs to be configured in the Hosts.
• VMs need to use be configured to use NUMA pinning and hugepages. See here.

Host Configuration

Set up Hugepages and iommu

Hugepages are virtual memory pages of a size greater than the 4K default. Increasing the size of the page reduces the number of pages in the system and hence the entries needed in the TLB to perform virtual address translations.

The size of virtual pages supported by the system can be check from the CPU flags:

• pse for 2M
• pdpe1gb for 1G

For 64-bit applications it is recommended to use 1G. Note that on NUMA systems, the pages reserved are divided equally between sockets.

For example, to configure default page size of 1G and 250 hugepages with iommu enabled at boot time on a Host with an Intel CPU, you have to append "intel_iommu=on default_hugepagesz=1G hugepagesz=1G hugepages=250" to the bootloader configuration.

# grep GRUB_CMDLINE_LINUX_DEFAULT /etc/default/grub
GRUB_CMDLINE_LINUX_DEFAULT="quiet splash intel_iommu=on default_hugepagesz=1G hugepagesz=1G hugepages=250"

# update-grub

Then reboot the system. After rebooting make sure that the hugepages mount can be seen so the applications can access them.

If you see the following, you don’t need to set up a mount as the mounts are already handled:

# mount | grep huge
hugetlbfs on /dev/hugepages type hugetlbfs (rw,relatime,pagesize=1G)
# systemctl list-units --type=mount | grep hugepages
dev-hugepages.mount                                    loaded active mounted Huge Pages File System

If not, you have to set it up yourself. To do this, create a mount point, for example, /mnt/hugepages1G and append nodev /mnt/hugepages1G hugetlbfs pagesize=1GB 0 0 as an entry to /etc/fstab

# mkdir /mnt/hugepages1G
# grep huge /etc/fstab
nodev       /mnt/hugepages1G hugetlbfs pagesize=1GB 0 0
# mount /mnt/hugepages1G

Now check hugepages are allocated to NUMA nodes, for example (or with numastat -m):

# grep -i '\<huge' /sys/devices/system/node/node*/meminfo
Node 0 HugePages_Total:   125
Node 0 HugePages_Free:    125
Node 0 HugePages_Surp:      0
Node 1 HugePages_Total:   125
Node 1 HugePages_Free:    125
Node 1 HugePages_Surp:      0

And finally iommu should be also enabled:

# dmesg | grep -i dmar
[    0.010651] kernel: ACPI: DMAR 0x000000007BAFE000 0000F0 (v01 DELL   PE_SC3   00000001 DELL 00000001)
[    0.010695] kernel: ACPI: Reserving DMAR table memory at [mem 0x7bafe000-0x7bafe0ef]
[    1.837579] kernel: DMAR: IOMMU enabled

Install OVS with DPDK support

We just need to install the dpdk version of the package and update alternatives accordingly:

# apt install openvswitch-switch-dpdk

# update-alternatives --set ovs-vswitchd /usr/lib/openvswitch-switch-dpdk/ovs-vswitchd-dpdk

# ovs-vsctl set Open_vSwitch . other_config:dpdk-init=true

Now, restart the openvswitch service and check dpdk is enabled:

# systemctl restart openvswitch-switch.service

# grep DPDK openvswitch/ovs-vswitchd.log
2022-11-24T12:30:24.500Z|00041|dpdk|ERR|DPDK not supported in this copy of Open vSwitch.
2022-11-24T12:33:02.905Z|00007|dpdk|INFO|Using DPDK 21.11.2
2022-11-24T12:33:02.905Z|00008|dpdk|INFO|DPDK Enabled - initializing...
2022-11-24T12:33:02.905Z|00012|dpdk|INFO|Per port memory for DPDK devices disabled.
2022-11-24T12:33:02.914Z|00016|dpdk|INFO|EAL: Detected shared linkage of DPDK
2022-11-24T12:33:04.303Z|00032|dpdk|INFO|DPDK Enabled - initialized

# ovs-vsctl get Open_vSwitch . dpdk_initialized
true

Configure Open vSwitch

The next step is to tune the execution parameters of the polling mode drivers (PMD) threads by pinning them into specific CPUs and assigning some hugepages.

To specify the CPU cores we need to set a binary mask, where each bit represents a CPU core by its ID. For example 0xF0 is 11110000, bits 7,6,5,4 are set to 1 so CPU cores 7,6,5,4 would be used for PMDs. Usually, it is recommended to allocate the same number of cores across NUMA nodes.

For example, to set cores 0,28,1,29 and 2G of hugepages per NUMA node on a Host with two sockets, execute the following commands:

# ovs-vsctl set Open_vSwitch . other_config:pmd-cpu-mask=0x30000003
# ovs-vsctl set Open_vSwitch . other_config:dpdk-socket-mem="2048,2048" # socket-mem=socket0_mem,socket1_mem
# ovs-vsctl set Open_vSwitch . other_config:dpdk-hugepage-dir="/mnt/hugepages1G"

# systemctl restart openvswitch-switch.service

Configure Open vSwitch Bridge

OpenNebula does not support adding and configuring DPDK physical devices. Binding cards to vfio-pci driver needs to be configured before using the DPDK network in OpenNebula. Usually, Open vSwitch setups only require one bridge so these steps can be easily automated during the Host installation.

In this example, we’ll create a bond with two cards (each attached to a different NUMA node). Let’s first trace the cards with the dpdk-debind.py tool and then bind the cards to the vfio-pci driver.

# dpdk-devbind.py --status
...
Network devices using kernel driver
===================================
0000:01:00.1 'Ethernet Controller X710 for 10GbE SFP+ 1572' if=eno2 drv=i40e unused=vfio-pci
0000:83:00.1 'Ethernet Controller X710 for 10GbE SFP+ 1572' if=enp131s0f1 drv=i40e unused=vfio-pci

# dpdk-devbind.py --bind=vfio-pci enp131s0f1

# dpdk-devbind.py --bind=vfio-pci eno2

# dpdk-devbind.py --status
...
Network devices using DPDK-compatible driver
============================================
0000:01:00.1 'Ethernet Controller X710 for 10GbE SFP+ 1572' drv=vfio-pci unused=i40e
0000:83:00.1 'Ethernet Controller X710 for 10GbE SFP+ 1572' drv=vfio-pci unused=i40e

Now we can add the cards to an Open vSwitch port, or in this example create a bond port with both:

# ovs-vsctl add-br onebr.dpdk -- set bridge onebr.dpdk datapath_type=netdev

# ovs-vsctl add-bond onebr.dpdk bond1 x710_1 x710_83 \
    -- set Interface x710_1 type=dpdk options:dpdk-devargs=0000:01:00.1 \
    -- set Interface x710_83 type=dpdk options:dpdk-devargs=0000:83:00.1

# ovs-vsctl show
   Bridge onebr.dpdk
        datapath_type: netdev
        Port onebr.dpdk
            Interface onebr.dpdk
                type: internal
        Port bond1
            Interface x710_83
                type: dpdk
                options: {dpdk-devargs="0000:83:00.1"}
            Interface x710_1
                type: dpdk
                options: {dpdk-devargs="0000:01:00.1"}
    ovs_version: "2.17.2"

Now we are all set! The bridge onebr.dpdk is ready to be used by OpenNebula.

OpenNebula Configuration

There are no special configuration on the OpenNebula server. Note that the sockets used by the vhost interface are created in the VM directory (/var/lib/one/datastores/<ds_id>/<vm_id>) and named after the switch port.

Using DPDK in your Virtual Networks

There are no additional changes, simply:

• Create your networks using the ovswitch driver, see above.
• Change configuration of the BRIDGE_TYPE of the network to openvswitch_dpdk using either the CLI command onevnet update or Sunstone.

An example of a Virtual Network template for the previous configuration could be:

NAME = "DPDK_VSBC_HA2"
BRIDGE = "onebr.dpdk"
BRIDGE_TYPE = "openvswitch_dpdk"
SECURITY_GROUPS = "0"
VLAN_ID = "1402"
VN_MAD = "ovswitch"

# note there is no PHYDEV, after creation it will show PHYDEV = ""

Using DPDK in your Virtual Machines

The following settings need to be enabled:

• Make sure that the NIC model is set to virtio. This setting can be added as a default in /etc/one/vmm_exec/vmm_exec_kvm.conf.
• In order to use the vhost-user interface in libvirt hugepages needs to be enabled. OVS reads/write network packages from/to the memory (hugepages) of the guest. The memory access mode MUST be shared and the VM MUST configure huge pages.

An example of a Virtual Machine template for the previous configuration could be:

NAME   = "DPDK_VM"
MEMORY = "4096"

NIC = [ NETWORK = "DPDK_VSBC_HA2" ]

TOPOLOGY = [
   CORES = "2",
   HUGEPAGE_SIZE = "1024",
   MEMORY_ACCESS = "shared",
   PIN_POLICY    = "THREAD",
   SOCKETS = "1",
   THREADS = "2"
]

You can verify that the VMs are using the vhost interface by looking at their domain definition in the Host. You should see something like:

<domain type='kvm' id='417'>
  <name>one-10</name>
  ...
  <devices>
    ...
    <interface type='vhostuser'>
      <mac address='02:00:c0:a8:7a:02'/>
      <source type='unix' path='/var/lib/one//datastores/0/10/one-10-0' mode='server'/>
      <target dev=''/>
      <model type='virtio'/>
      <alias name='net0'/>
      <address type='pci' domain='0x0000' bus='0x00' slot='0x03' function='0x0'/>
    </interface>
  ...
</domain>

And the associated port in the bridge using the qemu vhost interface:

Bridge onebr.dpdk
    datapath_type: netdev
    Port "one-10-0"
        tag: 1420
        Interface "one-10-0"
            type: dpdkvhostuserclient
            options: {vhost-server-path="/var/lib/one//datastores/0/10/one-10-0"}
...

Using Open vSwitch with Q-in-Q

Q-in-Q is an amendment to the IEEE 802.1Q specification that provides the capability for multiple VLAN tags to be inserted into a single Ethernet frame. Using Q-in-Q (aka C-VLAN, customer VLAN) tunneling allows us to create Layer 2 Ethernet connection between customers’ cloud infrastructure and OpenNebula VMs, or use a single service VLAN to bundle different customer VLANs.

OpenNebula Configuration

There is no configuration specific for this use case, just consider the general options specified above.

Defining a Q-in-Q Open vSwitch Network

To create a network you need to include the following information:

For example, you can define an Open vSwitch - QinQ Network with the following template:

NAME     = "qinq_net"
VN_MAD   = "ovswitch"
PHYDEV   = eth0
VLAN_ID  = 50                 # Service VLAN ID
CVLANS   = "101,103,110-113"  # Customer VLAN ID list

In this example, the driver will assign and create an Open vSwitch bridge and will attach the interface eth0 it. When a Virtual Machine is instantiated, its bridge ports will be tagged with 802.1Q VLAN ID 50 and service VLAN IDs 101,103,110,111,112,113. The configuration of the port should be similar to the that of following example that shows the second (NIC_ID=1) interface port one-1-5 for VM 5:

# ovs-vsctl list Port one-5-1

_uuid               : 791b84a9-2705-4cf9-94b4-43b39b98fe62
bond_active_slave   : []
bond_downdelay      : 0
bond_fake_iface     : false
bond_mode           : []
bond_updelay        : 0
cvlans              : [101, 103, 110, 111, 112, 113]
external_ids        : {}
fake_bridge         : false
interfaces          : [6da7ff07-51ec-40e9-97cd-c74a36e2c267]
lacp                : []
mac                 : []
name                : one-5-1
other_config        : {qinq-ethtype="802.1q"}
protected           : false
qos                 : []
rstp_statistics     : {}
rstp_status         : {}
statistics          : {}
status              : {}
tag                 : 100
trunks              : []
vlan_mode           : dot1q-tunnel

< Previous page