How to use Open vSwitch with DPDK

Since DPDK is just a library, it doesn’t do a lot on its own so it depends on emerging projects making use of it. One consumer of the library that is already part of Ubuntu is Open vSwitch with DPDK (OvS-DPDK) support in the package openvswitch-switch-dpdk.

Here is a brief example of how to install and configure a basic Open vSwitch using DPDK for later use via libvirt/qemu-kvm.

sudo apt-get install openvswitch-switch-dpdk
sudo update-alternatives --set ovs-vswitchd /usr/lib/openvswitch-switch-dpdk/ovs-vswitchd-dpdk
ovs-vsctl set Open_vSwitch . "other_config:dpdk-init=true"
# run on core 0 only
ovs-vsctl set Open_vSwitch . "other_config:dpdk-lcore-mask=0x1"
# Allocate 2G huge pages (not Numa node aware)
ovs-vsctl set Open_vSwitch . "other_config:dpdk-alloc-mem=2048"
# limit to one whitelisted device
ovs-vsctl set Open_vSwitch . "other_config:dpdk-extra=--pci-whitelist=0000:04:00.0"
sudo service openvswitch-switch restart

Remember: You need to assign devices to DPDK-compatible drivers before restarting – see the DPDK section on unassigning the default kernel drivers.

Please note that the section _dpdk-alloc-mem=2048_ in the above example is the most basic non-uniform memory access (NUMA) setup for a single socket system. If you have multiple sockets you may want to define how the memory should be split among them. More details about these options are outlined in Open vSwitch setup.

Attach DPDK ports to Open vSwitch

The Open vSwitch you started above supports all the same port types as Open vSwitch usually does, plus DPDK port types. The following example shows how to create a bridge and – instead of a normal external port – add an external DPDK port to it. When doing so you can specify the associated device.

ovs-vsctl add-br ovsdpdkbr0 -- set bridge ovsdpdkbr0 datapath_type=netdev
ovs-vsctl add-port ovsdpdkbr0 dpdk0 -- set Interface dpdk0 type=dpdk  "options:dpdk-devargs=${OVSDEV_PCIID}"

You can tune this further by setting options:

ovs-vsctl set Interface dpdk0 "options:n_rxq=2"

Open vSwitch DPDK to KVM guests

If you are not building some sort of software-defined networking (SDN) switch or NFV on top of DPDK, it is very likely that you want to forward traffic to KVM guests. The good news is; with the new qemu/libvirt/dpdk/openvswitch versions in Ubuntu this is no longer about manually appending a command line string. This section demonstrates a basic setup to connect a KVM guest to an Open vSwitch DPDK instance.

The recommended way to get to a KVM guest is using vhost_user_client. This will cause OvS-DPDK to connect to a socket created by QEMU. In this way, we can avoid old issues like “guest failures on OvS restart”. Here is an example of how to add such a port to the bridge you created above.

ovs-vsctl add-port ovsdpdkbr0 vhost-user-1 -- set Interface vhost-user-1 type=dpdkvhostuserclient "options:vhost-server-path=/var/run/vhostuserclient/vhost-user-client-1"

This will connect to the specified path that has to be created by a guest listening for it.

To let libvirt/kvm consume this socket and create a guest VirtIO network device for it, add the following snippet to your guest definition as the network definition.

<interface type='vhostuser'>
<source type='unix'
path='/var/run/vhostuserclient/vhost-user-client-1'
mode='server'/>
<model type='virtio'/>
</interface>

Tuning Open vSwitch-DPDK

DPDK has plenty of options – in combination with Open vSwitch-DPDK the two most commonly used are:

ovs-vsctl set Open_vSwitch . other_config:n-dpdk-rxqs=2
ovs-vsctl set Open_vSwitch . other_config:pmd-cpu-mask=0x6

The first line selects how many Rx Queues are to be used for each DPDK interface, while the second controls how many poll mode driver (PMD) threads to run (and where to run them). The example above will use two Rx Queues, and run PMD threads on CPU 1 and 2.

See also: Check the links to “EAL Command-line Options” and “Open vSwitch DPDK installation” at the end of this document for more information.

As usual with tunings, you need to know your system and workload really well - so please verify any tunings with workloads matching your real use case.

Support and troubleshooting

DPDK is a fast-evolving project. In any search for support and/or further guides, we highly recommended first checking to see if they apply to the current version.

You can check if your issues is known on:

Issues are often due to missing small details in the general setup. Later on, these missing details cause problems which can be hard to track down to their root cause.

A common case seems to be the “could not open network device dpdk0 (No such device)” issue. This occurs rather late when setting up a port in Open vSwitch with DPDK, but the root cause (most of the time) is very early in the setup and initialisation. Here is an example of how proper initialiasation of a device looks - this can be found in the syslog/journal when starting Open vSwitch with DPDK enabled.

ovs-ctl[3560]: EAL: PCI device 0000:04:00.1 on NUMA socket 0
ovs-ctl[3560]: EAL:   probe driver: 8086:1528 rte_ixgbe_pmd
ovs-ctl[3560]: EAL:   PCI memory mapped at 0x7f2140000000
ovs-ctl[3560]: EAL:   PCI memory mapped at 0x7f2140200000

If this is missing, either by ignored cards, failed initialisation or other reasons, later on there will be no DPDK device to refer to. Unfortunately, the logging is spread across syslog/journal and the openvswitch log. To enable some cross-checking, here is an example of what can be found in these logs, relative to the entered command.

#Note: This log was taken with dpdk 2.2 and openvswitch 2.5 but still looks quite similar (a bit extended) these days
Captions:
CMD: that you enter
SYSLOG: (Inlcuding EAL and OVS Messages)
OVS-LOG: (Openvswitch messages)

#PREPARATION
Bind an interface to DPDK UIO drivers, make Hugepages available, enable DPDK on OVS

CMD: sudo service openvswitch-switch restart

SYSLOG:
2016-01-22T08:58:31.372Z|00003|daemon_unix(monitor)|INFO|pid 3329 died, killed (Terminated), exiting
2016-01-22T08:58:33.377Z|00002|vlog|INFO|opened log file /var/log/openvswitch/ovs-vswitchd.log
2016-01-22T08:58:33.381Z|00003|ovs_numa|INFO|Discovered 12 CPU cores on NUMA node 0
2016-01-22T08:58:33.381Z|00004|ovs_numa|INFO|Discovered 1 NUMA nodes and 12 CPU cores
2016-01-22T08:58:33.381Z|00005|reconnect|INFO|unix:/var/run/openvswitch/db.sock: connecting...
2016-01-22T08:58:33.383Z|00006|reconnect|INFO|unix:/var/run/openvswitch/db.sock: connected
2016-01-22T08:58:33.386Z|00007|bridge|INFO|ovs-vswitchd (Open vSwitch) 2.5.0

OVS-LOG:
systemd[1]: Stopping Open vSwitch...
systemd[1]: Stopped Open vSwitch.
systemd[1]: Stopping Open vSwitch Internal Unit...
ovs-ctl[3541]: * Killing ovs-vswitchd (3329)
ovs-ctl[3541]: * Killing ovsdb-server (3318)
systemd[1]: Stopped Open vSwitch Internal Unit.
systemd[1]: Starting Open vSwitch Internal Unit...
ovs-ctl[3560]: * Starting ovsdb-server
ovs-vsctl: ovs|00001|vsctl|INFO|Called as ovs-vsctl --no-wait -- init -- set Open_vSwitch . db-version=7.12.1
ovs-vsctl: ovs|00001|vsctl|INFO|Called as ovs-vsctl --no-wait set Open_vSwitch . ovs-version=2.5.0 "external-ids:system-id=\"e7c5ba80-bb14-45c1-b8eb-628f3ad03903\"" "system-type=\"Ubuntu\"" "system-version=\"16.04-xenial\""
ovs-ctl[3560]: * Configuring Open vSwitch system IDs
ovs-ctl[3560]: 2016-01-22T08:58:31Z|00001|dpdk|INFO|No -vhost_sock_dir provided - defaulting to /var/run/openvswitch
ovs-vswitchd: ovs|00001|dpdk|INFO|No -vhost_sock_dir provided - defaulting to /var/run/openvswitch
ovs-ctl[3560]: EAL: Detected lcore 0 as core 0 on socket 0
ovs-ctl[3560]: EAL: Detected lcore 1 as core 1 on socket 0
ovs-ctl[3560]: EAL: Detected lcore 2 as core 2 on socket 0
ovs-ctl[3560]: EAL: Detected lcore 3 as core 3 on socket 0
ovs-ctl[3560]: EAL: Detected lcore 4 as core 4 on socket 0
ovs-ctl[3560]: EAL: Detected lcore 5 as core 5 on socket 0
ovs-ctl[3560]: EAL: Detected lcore 6 as core 0 on socket 0
ovs-ctl[3560]: EAL: Detected lcore 7 as core 1 on socket 0
ovs-ctl[3560]: EAL: Detected lcore 8 as core 2 on socket 0
ovs-ctl[3560]: EAL: Detected lcore 9 as core 3 on socket 0
ovs-ctl[3560]: EAL: Detected lcore 10 as core 4 on socket 0
ovs-ctl[3560]: EAL: Detected lcore 11 as core 5 on socket 0
ovs-ctl[3560]: EAL: Support maximum 128 logical core(s) by configuration.
ovs-ctl[3560]: EAL: Detected 12 lcore(s)
ovs-ctl[3560]: EAL: VFIO modules not all loaded, skip VFIO support...
ovs-ctl[3560]: EAL: Setting up physically contiguous memory...
ovs-ctl[3560]: EAL: Ask a virtual area of 0x100000000 bytes
ovs-ctl[3560]: EAL: Virtual area found at 0x7f2040000000 (size = 0x100000000)
ovs-ctl[3560]: EAL: Requesting 4 pages of size 1024MB from socket 0
ovs-ctl[3560]: EAL: TSC frequency is ~2397202 KHz
ovs-vswitchd[3592]: EAL: TSC frequency is ~2397202 KHz
ovs-vswitchd[3592]: EAL: Master lcore 0 is ready (tid=fc6cbb00;cpuset=[0])
ovs-vswitchd[3592]: EAL: PCI device 0000:04:00.0 on NUMA socket 0
ovs-vswitchd[3592]: EAL:   probe driver: 8086:1528 rte_ixgbe_pmd
ovs-vswitchd[3592]: EAL:   Not managed by a supported kernel driver, skipped
ovs-vswitchd[3592]: EAL: PCI device 0000:04:00.1 on NUMA socket 0
ovs-vswitchd[3592]: EAL:   probe driver: 8086:1528 rte_ixgbe_pmd
ovs-vswitchd[3592]: EAL:   PCI memory mapped at 0x7f2140000000
ovs-vswitchd[3592]: EAL:   PCI memory mapped at 0x7f2140200000
ovs-ctl[3560]: EAL: Master lcore 0 is ready (tid=fc6cbb00;cpuset=[0])
ovs-ctl[3560]: EAL: PCI device 0000:04:00.0 on NUMA socket 0
ovs-ctl[3560]: EAL:   probe driver: 8086:1528 rte_ixgbe_pmd
ovs-ctl[3560]: EAL:   Not managed by a supported kernel driver, skipped
ovs-ctl[3560]: EAL: PCI device 0000:04:00.1 on NUMA socket 0
ovs-ctl[3560]: EAL:   probe driver: 8086:1528 rte_ixgbe_pmd
ovs-ctl[3560]: EAL:   PCI memory mapped at 0x7f2140000000
ovs-ctl[3560]: EAL:   PCI memory mapped at 0x7f2140200000
ovs-vswitchd[3592]: PMD: eth_ixgbe_dev_init(): MAC: 4, PHY: 3
ovs-vswitchd[3592]: PMD: eth_ixgbe_dev_init(): port 0 vendorID=0x8086 deviceID=0x1528
ovs-ctl[3560]: PMD: eth_ixgbe_dev_init(): MAC: 4, PHY: 3
ovs-ctl[3560]: PMD: eth_ixgbe_dev_init(): port 0 vendorID=0x8086 deviceID=0x1528
ovs-ctl[3560]: Zone 0: name:<RG_MP_log_history>, phys:0x83fffdec0, len:0x2080, virt:0x7f213fffdec0, socket_id:0, flags:0
ovs-ctl[3560]: Zone 1: name:<MP_log_history>, phys:0x83fd73d40, len:0x28a0c0, virt:0x7f213fd73d40, socket_id:0, flags:0
ovs-ctl[3560]: Zone 2: name:<rte_eth_dev_data>, phys:0x83fd43380, len:0x2f700, virt:0x7f213fd43380, socket_id:0, flags:0
ovs-ctl[3560]: * Starting ovs-vswitchd
ovs-ctl[3560]: * Enabling remote OVSDB managers
systemd[1]: Started Open vSwitch Internal Unit.
systemd[1]: Starting Open vSwitch...
systemd[1]: Started Open vSwitch.


CMD: sudo ovs-vsctl add-br ovsdpdkbr0 -- set bridge ovsdpdkbr0 datapath_type=netdev

SYSLOG:
2016-01-22T08:58:56.344Z|00008|memory|INFO|37256 kB peak resident set size after 24.5 seconds
2016-01-22T08:58:56.346Z|00009|ofproto_dpif|INFO|netdev@ovs-netdev: Datapath supports recirculation
2016-01-22T08:58:56.346Z|00010|ofproto_dpif|INFO|netdev@ovs-netdev: MPLS label stack length probed as 3
2016-01-22T08:58:56.346Z|00011|ofproto_dpif|INFO|netdev@ovs-netdev: Datapath supports unique flow ids
2016-01-22T08:58:56.346Z|00012|ofproto_dpif|INFO|netdev@ovs-netdev: Datapath does not support ct_state
2016-01-22T08:58:56.346Z|00013|ofproto_dpif|INFO|netdev@ovs-netdev: Datapath does not support ct_zone
2016-01-22T08:58:56.346Z|00014|ofproto_dpif|INFO|netdev@ovs-netdev: Datapath does not support ct_mark
2016-01-22T08:58:56.346Z|00015|ofproto_dpif|INFO|netdev@ovs-netdev: Datapath does not support ct_label
2016-01-22T08:58:56.360Z|00016|bridge|INFO|bridge ovsdpdkbr0: added interface ovsdpdkbr0 on port 65534
2016-01-22T08:58:56.361Z|00017|bridge|INFO|bridge ovsdpdkbr0: using datapath ID 00005a4a1ed0a14d
2016-01-22T08:58:56.361Z|00018|connmgr|INFO|ovsdpdkbr0: added service controller "punix:/var/run/openvswitch/ovsdpdkbr0.mgmt"

OVS-LOG:
ovs-vsctl: ovs|00001|vsctl|INFO|Called as ovs-vsctl add-br ovsdpdkbr0 -- set bridge ovsdpdkbr0 datapath_type=netdev
systemd-udevd[3607]: Could not generate persistent MAC address for ovs-netdev: No such file or directory
kernel: [50165.886554] device ovs-netdev entered promiscuous mode
kernel: [50165.901261] device ovsdpdkbr0 entered promiscuous mode


CMD: sudo ovs-vsctl add-port ovsdpdkbr0 dpdk0 -- set Interface dpdk0 type=dpdk

SYSLOG:
2016-01-22T08:59:06.369Z|00019|memory|INFO|peak resident set size grew 155% in last 10.0 seconds, from 37256 kB to 95008 kB
2016-01-22T08:59:06.369Z|00020|memory|INFO|handlers:4 ports:1 revalidators:2 rules:5
2016-01-22T08:59:30.989Z|00021|dpdk|INFO|Port 0: 8c:dc:d4:b3:6d:e9
2016-01-22T08:59:31.520Z|00022|dpdk|INFO|Port 0: 8c:dc:d4:b3:6d:e9
2016-01-22T08:59:31.521Z|00023|dpif_netdev|INFO|Created 1 pmd threads on numa node 0
2016-01-22T08:59:31.522Z|00001|dpif_netdev(pmd16)|INFO|Core 0 processing port 'dpdk0'
2016-01-22T08:59:31.522Z|00024|bridge|INFO|bridge ovsdpdkbr0: added interface dpdk0 on port 1
2016-01-22T08:59:31.522Z|00025|bridge|INFO|bridge ovsdpdkbr0: using datapath ID 00008cdcd4b36de9
2016-01-22T08:59:31.523Z|00002|dpif_netdev(pmd16)|INFO|Core 0 processing port 'dpdk0'

OVS-LOG:
ovs-vsctl: ovs|00001|vsctl|INFO|Called as ovs-vsctl add-port ovsdpdkbr0 dpdk0 -- set Interface dpdk0 type=dpdk
ovs-vswitchd[3595]: PMD: ixgbe_dev_tx_queue_setup(): sw_ring=0x7f211a79ebc0 hw_ring=0x7f211a7a6c00 dma_addr=0x81a7a6c00
ovs-vswitchd[3595]: PMD: ixgbe_set_tx_function(): Using simple tx code path
ovs-vswitchd[3595]: PMD: ixgbe_set_tx_function(): Vector tx enabled.
ovs-vswitchd[3595]: PMD: ixgbe_dev_rx_queue_setup(): sw_ring=0x7f211a78a6c0 sw_sc_ring=0x7f211a786580 hw_ring=0x7f211a78e800 dma_addr=0x81a78e800
ovs-vswitchd[3595]: PMD: ixgbe_set_rx_function(): Vector rx enabled, please make sure RX burst size no less than 4 (port=0).
ovs-vswitchd[3595]: PMD: ixgbe_dev_tx_queue_setup(): sw_ring=0x7f211a79ebc0 hw_ring=0x7f211a7a6c00 dma_addr=0x81a7a6c00
...


CMD: sudo ovs-vsctl add-port ovsdpdkbr0 vhost-user-1 -- set Interface vhost-user-1 type=dpdkvhostuser

OVS-LOG:
2016-01-22T09:00:35.145Z|00026|dpdk|INFO|Socket /var/run/openvswitch/vhost-user-1 created for vhost-user port vhost-user-1
2016-01-22T09:00:35.145Z|00003|dpif_netdev(pmd16)|INFO|Core 0 processing port 'dpdk0'
2016-01-22T09:00:35.145Z|00004|dpif_netdev(pmd16)|INFO|Core 0 processing port 'vhost-user-1'
2016-01-22T09:00:35.145Z|00027|bridge|INFO|bridge ovsdpdkbr0: added interface vhost-user-1 on port 2

SYSLOG:
ovs-vsctl: ovs|00001|vsctl|INFO|Called as ovs-vsctl add-port ovsdpdkbr0 vhost-user-1 -- set Interface vhost-user-1 type=dpdkvhostuser
ovs-vswitchd[3595]: VHOST_CONFIG: socket created, fd:46
ovs-vswitchd[3595]: VHOST_CONFIG: bind to /var/run/openvswitch/vhost-user-1

Eventually we can see the poll thread in top
  PID USER      PR  NI    VIRT    RES    SHR S  %CPU %MEM     TIME+ COMMAND
 3595 root      10 -10 4975344 103936   9916 S 100.0  0.3  33:13.56 ovs-vswitchd

Resources