1 Using Open vSwitch with DPDK
2 ============================
4 Open vSwitch can use Intel(R) DPDK lib to operate entirely in
5 userspace. This file explains how to install and use Open vSwitch in
8 The DPDK support of Open vSwitch is considered experimental.
9 It has not been thoroughly tested.
11 This version of Open vSwitch should be built manually with `configure`
14 OVS needs a system with 1GB hugepages support.
16 Building and Installing:
17 ------------------------
20 Optional `fuse`, `fuse-devel`
22 1. Configure build & install DPDK:
26 export DPDK_DIR=/usr/src/dpdk-1.8.0
30 2. Update `config/common_linuxapp` so that DPDK generate single lib file.
31 (modification also required for IVSHMEM build)
33 `CONFIG_RTE_BUILD_COMBINE_LIBS=y`
35 Then run `make install` to build and isntall the library.
36 For default install without IVSHMEM:
38 `make install T=x86_64-native-linuxapp-gcc`
40 To include IVSHMEM (shared memory):
42 `make install T=x86_64-ivshmem-linuxapp-gcc`
44 For further details refer to http://dpdk.org/
46 2. Configure & build the Linux kernel:
48 Refer to intel-dpdk-getting-started-guide.pdf for understanding
49 DPDK kernel requirement.
51 3. Configure & build OVS:
55 `export DPDK_BUILD=$DPDK_DIR/x86_64-native-linuxapp-gcc/`
59 `export DPDK_BUILD=$DPDK_DIR/x86_64-ivshmem-linuxapp-gcc/`
62 cd $(OVS_DIR)/openvswitch
64 ./configure --with-dpdk=$DPDK_BUILD
68 To have better performance one can enable aggressive compiler optimizations and
69 use the special instructions(popcnt, crc32) that may not be available on all
70 machines. Instead of typing `make`, type:
72 `make CFLAGS='-O3 -march=native'`
74 Refer to [INSTALL.userspace.md] for general requirements of building userspace OVS.
76 Using the DPDK with ovs-vswitchd:
77 ---------------------------------
80 Add the following options to the kernel bootline:
82 `default_hugepagesz=1GB hugepagesz=1G hugepages=1`
84 2. Setup DPDK devices:
86 DPDK devices can be setup using either the VFIO (for DPDK 1.7+) or UIO
87 modules. UIO requires inserting an out of tree driver igb_uio.ko that is
88 available in DPDK. Setup for both methods are described below.
91 1. insert uio.ko: `modprobe uio`
92 2. insert igb_uio.ko: `insmod $DPDK_BUILD/kmod/igb_uio.ko`
93 3. Bind network device to igb_uio:
94 `$DPDK_DIR/tools/dpdk_nic_bind.py --bind=igb_uio eth1`
98 VFIO needs to be supported in the kernel and the BIOS. More information
99 can be found in the [DPDK Linux GSG].
101 1. Insert vfio-pci.ko: `modprobe vfio-pci`
102 2. Set correct permissions on vfio device: `sudo /usr/bin/chmod a+x /dev/vfio`
103 and: `sudo /usr/bin/chmod 0666 /dev/vfio/*`
104 3. Bind network device to vfio-pci:
105 `$DPDK_DIR/tools/dpdk_nic_bind.py --bind=vfio-pci eth1`
107 3. Mount the hugetable filsystem
109 `mount -t hugetlbfs -o pagesize=1G none /dev/hugepages`
111 Ref to http://www.dpdk.org/doc/quick-start for verifying DPDK setup.
113 4. Follow the instructions in [INSTALL.md] to install only the
114 userspace daemons and utilities (via 'make install').
115 1. First time only db creation (or clearing):
118 mkdir -p /usr/local/etc/openvswitch
119 mkdir -p /usr/local/var/run/openvswitch
120 rm /usr/local/etc/openvswitch/conf.db
121 ovsdb-tool create /usr/local/etc/openvswitch/conf.db \
122 /usr/local/share/openvswitch/vswitch.ovsschema
125 2. Start ovsdb-server
128 ovsdb-server --remote=punix:/usr/local/var/run/openvswitch/db.sock \
129 --remote=db:Open_vSwitch,Open_vSwitch,manager_options \
130 --private-key=db:Open_vSwitch,SSL,private_key \
131 --certificate=Open_vSwitch,SSL,certificate \
132 --bootstrap-ca-cert=db:Open_vSwitch,SSL,ca_cert --pidfile --detach
135 3. First time after db creation, initialize:
138 ovs-vsctl --no-wait init
143 DPDK configuration arguments can be passed to vswitchd via `--dpdk`
144 argument. This needs to be first argument passed to vswitchd process.
145 dpdk arg -c is ignored by ovs-dpdk, but it is a required parameter
146 for dpdk initialization.
149 export DB_SOCK=/usr/local/var/run/openvswitch/db.sock
150 ovs-vswitchd --dpdk -c 0x1 -n 4 -- unix:$DB_SOCK --pidfile --detach
153 If allocated more than one GB hugepage (as for IVSHMEM), set amount and
154 use NUMA node 0 memory:
157 ovs-vswitchd --dpdk -c 0x1 -n 4 --socket-mem 1024,0 \
158 -- unix:$DB_SOCK --pidfile --detach
161 6. Add bridge & ports
163 To use ovs-vswitchd with DPDK, create a bridge with datapath_type
164 "netdev" in the configuration database. For example:
166 `ovs-vsctl add-br br0 -- set bridge br0 datapath_type=netdev`
168 Now you can add dpdk devices. OVS expect DPDK device name start with dpdk
169 and end with portid. vswitchd should print (in the log file) the number
170 of dpdk devices found.
173 ovs-vsctl add-port br0 dpdk0 -- set Interface dpdk0 type=dpdk
174 ovs-vsctl add-port br0 dpdk1 -- set Interface dpdk1 type=dpdk
177 Once first DPDK port is added to vswitchd, it creates a Polling thread and
178 polls dpdk device in continuous loop. Therefore CPU utilization
179 for that thread is always 100%.
183 Test flow script across NICs (assuming ovs in /usr/src/ovs):
188 # Move to command directory
189 cd /usr/src/ovs/utilities/
191 # Clear current flows
192 ./ovs-ofctl del-flows br0
194 # Add flows between port 1 (dpdk0) to port 2 (dpdk1)
195 ./ovs-ofctl add-flow br0 in_port=1,action=output:2
196 ./ovs-ofctl add-flow br0 in_port=2,action=output:1
199 8. Performance tuning
201 With pmd multi-threading support, OVS creates one pmd thread for each
202 numa node as default. The pmd thread handles the I/O of all DPDK
203 interfaces on the same numa node. The following two commands can be used
204 to configure the multi-threading behavior.
206 `ovs-vsctl set Open_vSwitch . other_config:pmd-cpu-mask=<hex string>`
208 The command above asks for a CPU mask for setting the affinity of pmd
209 threads. A set bit in the mask means a pmd thread is created and pinned
210 to the corresponding CPU core. For more information, please refer to
211 `man ovs-vswitchd.conf.db`
213 `ovs-vsctl set Open_vSwitch . other_config:n-dpdk-rxqs=<integer>`
215 The command above sets the number of rx queues of each DPDK interface. The
216 rx queues are assigned to pmd threads on the same numa node in round-robin
217 fashion. For more information, please refer to `man ovs-vswitchd.conf.db`
219 Ideally for maximum throughput, the pmd thread should not be scheduled out
220 which temporarily halts its execution. The following affinitization methods
223 Lets pick core 4,6,8,10 for pmd threads to run on. Also assume a dual 8 core
224 sandy bridge system with hyperthreading enabled where CPU1 has cores 0,...,7
225 and 16,...,23 & CPU2 cores 8,...,15 & 24,...,31. (A different cpu
226 configuration could have different core mask requirements).
228 To kernel bootline add core isolation list for cores and associated hype cores
229 (e.g. isolcpus=4,20,6,22,8,24,10,26,). Reboot system for isolation to take
230 effect, restart everything.
232 Configure pmd threads on core 4,6,8,10 using 'pmd-cpu-mask':
234 `ovs-vsctl set Open_vSwitch . other_config:pmd-cpu-mask=00000550`
236 You should be able to check that pmd threads are pinned to the correct cores
240 top -p `pidof ovs-vswitchd` -H -d1
243 Note, the pmd threads on a numa node are only created if there is at least
244 one DPDK interface from the numa node that has been added to OVS.
246 Note, core 0 is always reserved from non-pmd threads and should never be set
252 Following the steps above to create a bridge, you can now add dpdk rings
253 as a port to the vswitch. OVS will expect the DPDK ring device name to
254 start with dpdkr and end with a portid.
256 `ovs-vsctl add-port br0 dpdkr0 -- set Interface dpdkr0 type=dpdkr`
258 DPDK rings client test application
260 Included in the test directory is a sample DPDK application for testing
261 the rings. This is from the base dpdk directory and modified to work
262 with the ring naming used within ovs.
264 location tests/ovs_client
269 cd /usr/src/ovs/tests/
270 ovsclient -c 1 -n 4 --proc-type=secondary -- -n "port id you gave dpdkr"
273 In the case of the dpdkr example above the "port id you gave dpdkr" is 0.
275 It is essential to have --proc-type=secondary
277 The application simply receives an mbuf on the receive queue of the
278 ethernet ring and then places that same mbuf on the transmit ring of
279 the ethernet ring. It is a trivial loopback application.
281 DPDK rings in VM (IVSHMEM shared memory communications)
282 -------------------------------------------------------
284 In addition to executing the client in the host, you can execute it within
285 a guest VM. To do so you will need a patched qemu. You can download the
286 patch and getting started guide at :
288 https://01.org/packet-processing/downloads
290 A general rule of thumb for better performance is that the client
291 application should not be assigned the same dpdk core mask "-c" as
297 vhost-cuse is only supported at present i.e. not using the standard QEMU
298 vhost-user interface. It is intended that vhost-user support will be added
299 in future releases when supported in DPDK and that vhost-cuse will eventually
300 be deprecated. See [DPDK Docs] for more info on vhost.
303 1. DPDK 1.8 with vhost support enabled and recompile OVS as above.
305 Update `config/common_linuxapp` so that DPDK is built with vhost
308 `CONFIG_RTE_LIBRTE_VHOST=y`
310 2. Insert the Cuse module:
314 3. Build and insert the `eventfd_link` module:
316 `cd $DPDK_DIR/lib/librte_vhost/eventfd_link/`
318 `insmod $DPDK_DIR/lib/librte_vhost/eventfd_link.ko`
320 Following the steps above to create a bridge, you can now add DPDK vhost
321 as a port to the vswitch.
323 `ovs-vsctl add-port br0 dpdkvhost0 -- set Interface dpdkvhost0 type=dpdkvhost`
325 Unlike DPDK ring ports, DPDK vhost ports can have arbitrary names:
327 `ovs-vsctl add-port br0 port123ABC -- set Interface port123ABC type=dpdkvhost`
329 However, please note that when attaching userspace devices to QEMU, the
330 name provided during the add-port operation must match the ifname parameter
331 on the QEMU command line.
334 DPDK vhost VM configuration:
335 ----------------------------
337 vhost ports use a Linux* character device to communicate with QEMU.
338 By default it is set to `/dev/vhost-net`. It is possible to reuse this
339 standard device for DPDK vhost, which makes setup a little simpler but it
340 is better practice to specify an alternative character device in order to
341 avoid any conflicts if kernel vhost is to be used in parallel.
343 1. This step is only needed if using an alternative character device.
345 The new character device filename must be specified on the vswitchd
348 `./vswitchd/ovs-vswitchd --dpdk --cuse_dev_name my-vhost-net -c 0x1 ...`
350 Note that the `--cuse_dev_name` argument and associated string must be the first
351 arguments after `--dpdk` and come before the EAL arguments. In the example
352 above, the character device to be used will be `/dev/my-vhost-net`.
354 2. This step is only needed if reusing the standard character device. It will
355 conflict with the kernel vhost character device so the user must first
358 `rm -rf /dev/vhost-net`
360 3a. Configure virtio-net adaptors:
361 The following parameters must be passed to the QEMU binary:
364 -netdev tap,id=<id>,script=no,downscript=no,ifname=<name>,vhost=on
365 -device virtio-net-pci,netdev=net1,mac=<mac>
368 Repeat the above parameters for multiple devices.
370 The DPDK vhost library will negiotiate its own features, so they
371 need not be passed in as command line params. Note that as offloads are
372 disabled this is the equivalent of setting:
374 `csum=off,gso=off,guest_tso4=off,guest_tso6=off,guest_ecn=off`
376 3b. If using an alternative character device. It must be also explicitly
377 passed to QEMU using the `vhostfd` argument:
380 -netdev tap,id=<id>,script=no,downscript=no,ifname=<name>,vhost=on,
382 -device virtio-net-pci,netdev=net1,mac=<mac>
385 The open file descriptor must be passed to QEMU running as a child
386 process. This could be done with a simple python script.
390 fd = os.open("/dev/usvhost", os.O_RDWR)
391 subprocess.call("qemu-system-x86_64 .... -netdev tap,id=vhostnet0,\
392 vhost=on,vhostfd=" + fd +"...", shell=True)
394 Alternatively the the `qemu-wrap.py` script can be used to automate the
395 requirements specified above and can be used in conjunction with libvirt if
396 desired. See the "DPDK vhost VM configuration with QEMU wrapper" section
399 4. Configure huge pages:
400 QEMU must allocate the VM's memory on hugetlbfs. Vhost ports access a
401 virtio-net device's virtual rings and packet buffers mapping the VM's
402 physical memory on hugetlbfs. To enable vhost-ports to map the VM's
403 memory into their process address space, pass the following paramters
406 `-object memory-backend-file,id=mem,size=4096M,mem-path=/dev/hugepages,
407 share=on -numa node,memdev=mem -mem-prealloc`
410 DPDK vhost VM configuration with QEMU wrapper:
411 ----------------------------------------------
413 The QEMU wrapper script automatically detects and calls QEMU with the
414 necessary parameters. It performs the following actions:
416 * Automatically detects the location of the hugetlbfs and inserts this
417 into the command line parameters.
418 * Automatically open file descriptors for each virtio-net device and
419 inserts this into the command line parameters.
420 * Calls QEMU passing both the command line parameters passed to the
421 script itself and those it has auto-detected.
423 Before use, you **must** edit the configuration parameters section of the
424 script to point to the correct emulator location and set additional
425 settings. Of these settings, `emul_path` and `us_vhost_path` **must** be
426 set. All other settings are optional.
428 To use directly from the command line simply pass the wrapper some of the
429 QEMU parameters: it will configure the rest. For example:
432 qemu-wrap.py -cpu host -boot c -hda <disk image> -m 4096 -smp 4
433 --enable-kvm -nographic -vnc none -net none -netdev tap,id=net1,
434 script=no,downscript=no,ifname=if1,vhost=on -device virtio-net-pci,
435 netdev=net1,mac=00:00:00:00:00:01
437 DPDK vhost VM configuration with libvirt:
438 -----------------------------------------
440 If you are using libvirt, you must enable libvirt to access the character
441 device by adding it to controllers cgroup for libvirtd using the following
444 1. In `/etc/libvirt/qemu.conf` add/edit the following lines:
447 1) clear_emulator_capabilities = 0
450 4) cgroup_device_acl = [
451 "/dev/null", "/dev/full", "/dev/zero",
452 "/dev/random", "/dev/urandom",
453 "/dev/ptmx", "/dev/kvm", "/dev/kqemu",
454 "/dev/rtc", "/dev/hpet", "/dev/net/tun",
455 "/dev/<my-vhost-device>",
459 <my-vhost-device> refers to "vhost-net" if using the `/dev/vhost-net`
460 device. If you have specificed a different name on the ovs-vswitchd
461 commandline using the "--cuse_dev_name" parameter, please specify that
464 2. Disable SELinux or set to permissive mode
466 3. Restart the libvirtd process
467 For example, on Fedora:
469 `systemctl restart libvirtd.service`
471 After successfully editing the configuration, you may launch your
472 vhost-enabled VM. The XML describing the VM can be configured like so
473 within the <qemu:commandline> section:
475 1. Set up shared hugepages:
478 <qemu:arg value='-object'/>
479 <qemu:arg value='memory-backend-file,id=mem,size=4096M,mem-path=/dev/hugepages,share=on'/>
480 <qemu:arg value='-numa'/>
481 <qemu:arg value='node,memdev=mem'/>
482 <qemu:arg value='-mem-prealloc'/>
485 2. Set up your tap devices:
488 <qemu:arg value='-netdev'/>
489 <qemu:arg value='type=tap,id=net1,script=no,downscript=no,ifname=vhost0,vhost=on'/>
490 <qemu:arg value='-device'/>
491 <qemu:arg value='virtio-net-pci,netdev=net1,mac=00:00:00:00:00:01'/>
494 Repeat for as many devices as are desired, modifying the id, ifname
495 and mac as necessary.
497 Again, if you are using an alternative character device (other than
498 `/dev/vhost-net`), please specify the file descriptor like so:
500 `<qemu:arg value='type=tap,id=net3,script=no,downscript=no,ifname=vhost0,vhost=on,vhostfd=<open_fd>'/>`
502 Where <open_fd> refers to the open file descriptor of the character device.
503 Instructions of how to retrieve the file descriptor can be found in the
504 "DPDK vhost VM configuration" section.
505 Alternatively, the process is automated with the qemu-wrap.py script,
506 detailed in the next section.
508 Now you may launch your VM using virt-manager, or like so:
510 `virsh create my_vhost_vm.xml`
512 DPDK vhost VM configuration with libvirt and QEMU wrapper:
513 ----------------------------------------------------------
515 To use the qemu-wrapper script in conjuntion with libvirt, follow the
516 steps in the previous section before proceeding with the following steps:
518 1. Place `qemu-wrap.py` in libvirtd's binary search PATH ($PATH)
519 Ideally in the same directory that the QEMU binary is located.
521 2. Ensure that the script has the same owner/group and file permissions
524 3. Update the VM xml file using "virsh edit VM.xml"
526 1. Set the VM to use the launch script.
527 Set the emulator path contained in the `<emulator><emulator/>` tags.
528 For example, replace:
530 `<emulator>/usr/bin/qemu-kvm<emulator/>`
534 `<emulator>/usr/bin/qemu-wrap.py<emulator/>`
536 4. Edit the Configuration Parameters section of the script to point to
537 the correct emulator location and set any additional options. If you are
538 using a alternative character device name, please set "us_vhost_path" to the
539 location of that device. The script will automatically detect and insert
540 the correct "vhostfd" value in the QEMU command line arguements.
542 5. Use virt-manager to launch the VM
547 - Work with 1500 MTU, needs few changes in DPDK lib to fix this issue.
548 - Currently DPDK port does not make use any offload functionality.
549 - DPDK-vHost support works with 1G huge pages.
552 - The shared memory is currently restricted to the use of a 1GB
554 - All huge pages are shared amongst the host, clients, virtual
560 Please report problems to bugs@openvswitch.org.
562 [INSTALL.userspace.md]:INSTALL.userspace.md
563 [INSTALL.md]:INSTALL.md
564 [DPDK Linux GSG]: http://www.dpdk.org/doc/guides/linux_gsg/build_dpdk.html#binding-and-unbinding-network-ports-to-from-the-igb-uioor-vfio-modules
565 [DPDK Docs]: http://dpdk.org/doc