1 .. SPDX-License-Identifier: BSD-3-Clause
2 Copyright(c) 2010-2015 Intel Corporation.
4 Link Bonding Poll Mode Driver Library
5 =====================================
7 In addition to Poll Mode Drivers (PMDs) for physical and virtual hardware,
8 DPDK also includes a pure-software library that
9 allows physical PMDs to be bonded together to create a single logical PMD.
11 .. figure:: img/bond-overview.*
16 The Link Bonding PMD library(librte_pmd_bond) supports bonding of groups of
17 ``rte_eth_dev`` ports of the same speed and duplex to provide similar
18 capabilities to that found in Linux bonding driver to allow the aggregation
19 of multiple (slave) NICs into a single logical interface between a server
20 and a switch. The new bonded PMD will then process these interfaces based on
21 the mode of operation specified to provide support for features such as
22 redundant links, fault tolerance and/or load balancing.
24 The librte_pmd_bond library exports a C API which provides an API for the
25 creation of bonded devices as well as the configuration and management of the
26 bonded device and its slave devices.
30 The Link Bonding PMD Library is enabled by default in the build
31 configuration files, the library can be disabled by setting
32 ``CONFIG_RTE_LIBRTE_PMD_BOND=n`` and recompiling the DPDK.
34 Link Bonding Modes Overview
35 ---------------------------
37 Currently the Link Bonding PMD library supports following modes of operation:
39 * **Round-Robin (Mode 0):**
41 .. figure:: img/bond-mode-0.*
46 This mode provides load balancing and fault tolerance by transmission of
47 packets in sequential order from the first available slave device through
48 the last. Packets are bulk dequeued from devices then serviced in a
49 round-robin manner. This mode does not guarantee in order reception of
50 packets and down stream should be able to handle out of order packets.
52 * **Active Backup (Mode 1):**
54 .. figure:: img/bond-mode-1.*
56 Active Backup (Mode 1)
59 In this mode only one slave in the bond is active at any time, a different
60 slave becomes active if, and only if, the primary active slave fails,
61 thereby providing fault tolerance to slave failure. The single logical
62 bonded interface's MAC address is externally visible on only one NIC (port)
63 to avoid confusing the network switch.
65 * **Balance XOR (Mode 2):**
67 .. figure:: img/bond-mode-2.*
72 This mode provides transmit load balancing (based on the selected
73 transmission policy) and fault tolerance. The default policy (layer2) uses
74 a simple calculation based on the packet flow source and destination MAC
75 addresses as well as the number of active slaves available to the bonded
76 device to classify the packet to a specific slave to transmit on. Alternate
77 transmission policies supported are layer 2+3, this takes the IP source and
78 destination addresses into the calculation of the transmit slave port and
79 the final supported policy is layer 3+4, this uses IP source and
80 destination addresses as well as the TCP/UDP source and destination port.
83 The coloring differences of the packets are used to identify different flow
84 classification calculated by the selected transmit policy
87 * **Broadcast (Mode 3):**
89 .. figure:: img/bond-mode-3.*
94 This mode provides fault tolerance by transmission of packets on all slave
97 * **Link Aggregation 802.3AD (Mode 4):**
99 .. figure:: img/bond-mode-4.*
101 Link Aggregation 802.3AD (Mode 4)
104 This mode provides dynamic link aggregation according to the 802.3ad
105 specification. It negotiates and monitors aggregation groups that share the
106 same speed and duplex settings using the selected balance transmit policy
107 for balancing outgoing traffic.
109 DPDK implementation of this mode provide some additional requirements of
112 #. It needs to call ``rte_eth_tx_burst`` and ``rte_eth_rx_burst`` with
113 intervals period of less than 100ms.
115 #. Calls to ``rte_eth_tx_burst`` must have a buffer size of at least 2xN,
116 where N is the number of slaves. This is a space required for LACP
117 frames. Additionally LACP packets are included in the statistics, but
118 they are not returned to the application.
120 * **Transmit Load Balancing (Mode 5):**
122 .. figure:: img/bond-mode-5.*
124 Transmit Load Balancing (Mode 5)
127 This mode provides an adaptive transmit load balancing. It dynamically
128 changes the transmitting slave, according to the computed load. Statistics
129 are collected in 100ms intervals and scheduled every 10ms.
132 Implementation Details
133 ----------------------
135 The librte_pmd_bond bonded device are compatible with the Ethernet device API
136 exported by the Ethernet PMDs described in the *DPDK API Reference*.
138 The Link Bonding Library supports the creation of bonded devices at application
139 startup time during EAL initialization using the ``--vdev`` option as well as
140 programmatically via the C API ``rte_eth_bond_create`` function.
142 Bonded devices support the dynamical addition and removal of slave devices using
143 the ``rte_eth_bond_slave_add`` / ``rte_eth_bond_slave_remove`` APIs.
145 After a slave device is added to a bonded device slave is stopped using
146 ``rte_eth_dev_stop`` and then reconfigured using ``rte_eth_dev_configure``
147 the RX and TX queues are also reconfigured using ``rte_eth_tx_queue_setup`` /
148 ``rte_eth_rx_queue_setup`` with the parameters use to configure the bonding
149 device. If RSS is enabled for bonding device, this mode is also enabled on new
150 slave and configured as well.
151 Any flow which was configured to the bond device also is configured to the added
154 Setting up multi-queue mode for bonding device to RSS, makes it fully
155 RSS-capable, so all slaves are synchronized with its configuration. This mode is
156 intended to provide RSS configuration on slaves transparent for client
157 application implementation.
159 Bonding device stores its own version of RSS settings i.e. RETA, RSS hash
160 function and RSS key, used to set up its slaves. That let to define the meaning
161 of RSS configuration of bonding device as desired configuration of whole bonding
162 (as one unit), without pointing any of slave inside. It is required to ensure
163 consistency and made it more error-proof.
165 RSS hash function set for bonding device, is a maximal set of RSS hash functions
166 supported by all bonded slaves. RETA size is a GCD of all its RETA's sizes, so
167 it can be easily used as a pattern providing expected behavior, even if slave
168 RETAs' sizes are different. If RSS Key is not set for bonded device, it's not
169 changed on the slaves and default key for device is used.
171 As RSS configurations, there is flow consistency in the bonded slaves for the
172 next rte flow operations:
175 - Validate flow for each slave, failure at least for one slave causes to
176 bond validation failure.
179 - Create the flow in all slaves.
180 - Save all the slaves created flows objects in bonding internal flow
182 - Failure in flow creation for existed slave rejects the flow.
183 - Failure in flow creation for new slaves in slave adding time rejects
187 - Destroy the flow in all slaves and release the bond internal flow
191 - Destroy all the bonding PMD flows in all the slaves.
195 Don't call slaves flush directly, It destroys all the slave flows which
196 may include external flows or the bond internal LACP flow.
199 - Summarize flow counters from all the slaves, relevant only for
200 ``RTE_FLOW_ACTION_TYPE_COUNT``.
203 - Call to flow isolate for all slaves.
204 - Failure in flow isolation for existed slave rejects the isolate mode.
205 - Failure in flow isolation for new slaves in slave adding time rejects
208 All settings are managed through the bonding port API and always are propagated
209 in one direction (from bonding to slaves).
211 Link Status Change Interrupts / Polling
212 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
214 Link bonding devices support the registration of a link status change callback,
215 using the ``rte_eth_dev_callback_register`` API, this will be called when the
216 status of the bonding device changes. For example in the case of a bonding
217 device which has 3 slaves, the link status will change to up when one slave
218 becomes active or change to down when all slaves become inactive. There is no
219 callback notification when a single slave changes state and the previous
220 conditions are not met. If a user wishes to monitor individual slaves then they
221 must register callbacks with that slave directly.
223 The link bonding library also supports devices which do not implement link
224 status change interrupts, this is achieved by polling the devices link status at
225 a defined period which is set using the ``rte_eth_bond_link_monitoring_set``
226 API, the default polling interval is 10ms. When a device is added as a slave to
227 a bonding device it is determined using the ``RTE_PCI_DRV_INTR_LSC`` flag
228 whether the device supports interrupts or whether the link status should be
229 monitored by polling it.
231 Requirements / Limitations
232 ~~~~~~~~~~~~~~~~~~~~~~~~~~
234 The current implementation only supports devices that support the same speed
235 and duplex to be added as a slaves to the same bonded device. The bonded device
236 inherits these attributes from the first active slave added to the bonded
237 device and then all further slaves added to the bonded device must support
240 A bonding device must have a minimum of one slave before the bonding device
241 itself can be started.
243 To use a bonding device dynamic RSS configuration feature effectively, it is
244 also required, that all slaves should be RSS-capable and support, at least one
245 common hash function available for each of them. Changing RSS key is only
246 possible, when all slave devices support the same key size.
248 To prevent inconsistency on how slaves process packets, once a device is added
249 to a bonding device, RSS and rte flow configurations should be managed through
250 the bonding device API, and not directly on the slave.
252 Like all other PMD, all functions exported by a PMD are lock-free functions
253 that are assumed not to be invoked in parallel on different logical cores to
254 work on the same target object.
256 It should also be noted that the PMD receive function should not be invoked
257 directly on a slave devices after they have been to a bonded device since
258 packets read directly from the slave device will no longer be available to the
259 bonded device to read.
264 Link bonding devices are created using the ``rte_eth_bond_create`` API
265 which requires a unique device name, the bonding mode,
266 and the socket Id to allocate the bonding device's resources on.
267 The other configurable parameters for a bonded device are its slave devices,
268 its primary slave, a user defined MAC address and transmission policy to use if
269 the device is in balance XOR mode.
274 Bonding devices support up to a maximum of ``RTE_MAX_ETHPORTS`` slave devices
275 of the same speed and duplex. Ethernet devices can be added as a slave to a
276 maximum of one bonded device. Slave devices are reconfigured with the
277 configuration of the bonded device on being added to a bonded device.
279 The bonded also guarantees to return the MAC address of the slave device to its
280 original value of removal of a slave from it.
285 The primary slave is used to define the default port to use when a bonded
286 device is in active backup mode. A different port will only be used if, and
287 only if, the current primary port goes down. If the user does not specify a
288 primary port it will default to being the first port added to the bonded device.
293 The bonded device can be configured with a user specified MAC address, this
294 address will be inherited by the some/all slave devices depending on the
295 operating mode. If the device is in active backup mode then only the primary
296 device will have the user specified MAC, all other slaves will retain their
297 original MAC address. In mode 0, 2, 3, 4 all slaves devices are configure with
298 the bonded devices MAC address.
300 If a user defined MAC address is not defined then the bonded device will
301 default to using the primary slaves MAC address.
303 Balance XOR Transmit Policies
304 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
306 There are 3 supported transmission policies for bonded device running in
307 Balance XOR mode. Layer 2, Layer 2+3, Layer 3+4.
309 * **Layer 2:** Ethernet MAC address based balancing is the default
310 transmission policy for Balance XOR bonding mode. It uses a simple XOR
311 calculation on the source MAC address and destination MAC address of the
312 packet and then calculate the modulus of this value to calculate the slave
313 device to transmit the packet on.
315 * **Layer 2 + 3:** Ethernet MAC address & IP Address based balancing uses a
316 combination of source/destination MAC addresses and the source/destination
317 IP addresses of the data packet to decide which slave port the packet will
320 * **Layer 3 + 4:** IP Address & UDP Port based balancing uses a combination
321 of source/destination IP Address and the source/destination UDP ports of
322 the packet of the data packet to decide which slave port the packet will be
325 All these policies support 802.1Q VLAN Ethernet packets, as well as IPv4, IPv6
326 and UDP protocols for load balancing.
328 Using Link Bonding Devices
329 --------------------------
331 The librte_pmd_bond library supports two modes of device creation, the libraries
332 export full C API or using the EAL command line to statically configure link
333 bonding devices at application startup. Using the EAL option it is possible to
334 use link bonding functionality transparently without specific knowledge of the
335 libraries API, this can be used, for example, to add bonding functionality,
336 such as active backup, to an existing application which has no knowledge of
337 the link bonding C API.
339 Using the Poll Mode Driver from an Application
340 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
342 Using the librte_pmd_bond libraries API it is possible to dynamically create
343 and manage link bonding device from within any application. Link bonding
344 devices are created using the ``rte_eth_bond_create`` API which requires a
345 unique device name, the link bonding mode to initial the device in and finally
346 the socket Id which to allocate the devices resources onto. After successful
347 creation of a bonding device it must be configured using the generic Ethernet
348 device configure API ``rte_eth_dev_configure`` and then the RX and TX queues
349 which will be used must be setup using ``rte_eth_tx_queue_setup`` /
350 ``rte_eth_rx_queue_setup``.
352 Slave devices can be dynamically added and removed from a link bonding device
353 using the ``rte_eth_bond_slave_add`` / ``rte_eth_bond_slave_remove``
354 APIs but at least one slave device must be added to the link bonding device
355 before it can be started using ``rte_eth_dev_start``.
357 The link status of a bonded device is dictated by that of its slaves, if all
358 slave device link status are down or if all slaves are removed from the link
359 bonding device then the link status of the bonding device will go down.
361 It is also possible to configure / query the configuration of the control
362 parameters of a bonded device using the provided APIs
363 ``rte_eth_bond_mode_set/ get``, ``rte_eth_bond_primary_set/get``,
364 ``rte_eth_bond_mac_set/reset`` and ``rte_eth_bond_xmit_policy_set/get``.
366 Using Link Bonding Devices from the EAL Command Line
367 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
369 Link bonding devices can be created at application startup time using the
370 ``--vdev`` EAL command line option. The device name must start with the
371 net_bonding prefix followed by numbers or letters. The name must be unique for
372 each device. Each device can have multiple options arranged in a comma
373 separated list. Multiple devices definitions can be arranged by calling the
374 ``--vdev`` option multiple times.
376 Device names and bonding options must be separated by commas as shown below:
378 .. code-block:: console
380 $RTE_TARGET/app/testpmd -l 0-3 -n 4 --vdev 'net_bonding0,bond_opt0=..,bond opt1=..'--vdev 'net_bonding1,bond _opt0=..,bond_opt1=..'
382 Link Bonding EAL Options
383 ^^^^^^^^^^^^^^^^^^^^^^^^
385 There are multiple ways of definitions that can be assessed and combined as
386 long as the following two rules are respected:
388 * A unique device name, in the format of net_bondingX is provided,
389 where X can be any combination of numbers and/or letters,
390 and the name is no greater than 32 characters long.
392 * A least one slave device is provided with for each bonded device definition.
394 * The operation mode of the bonded device being created is provided.
396 The different options are:
398 * mode: Integer value defining the bonding mode of the device.
399 Currently supports modes 0,1,2,3,4,5 (round-robin, active backup, balance,
400 broadcast, link aggregation, transmit load balancing).
402 .. code-block:: console
406 * slave: Defines the PMD device which will be added as slave to the bonded
407 device. This option can be selected multiple times, for each device to be
408 added as a slave. Physical devices should be specified using their PCI
409 address, in the format domain:bus:devid.function
411 .. code-block:: console
413 slave=0000:0a:00.0,slave=0000:0a:00.1
415 * primary: Optional parameter which defines the primary slave port,
416 is used in active backup mode to select the primary slave for data TX/RX if
417 it is available. The primary port also is used to select the MAC address to
418 use when it is not defined by the user. This defaults to the first slave
419 added to the device if it is specified. The primary device must be a slave
420 of the bonded device.
422 .. code-block:: console
426 * socket_id: Optional parameter used to select which socket on a NUMA device
427 the bonded devices resources will be allocated on.
429 .. code-block:: console
433 * mac: Optional parameter to select a MAC address for link bonding device,
434 this overrides the value of the primary slave device.
436 .. code-block:: console
438 mac=00:1e:67:1d:fd:1d
440 * xmit_policy: Optional parameter which defines the transmission policy when
441 the bonded device is in balance mode. If not user specified this defaults
442 to l2 (layer 2) forwarding, the other transmission policies available are
443 l23 (layer 2+3) and l34 (layer 3+4)
445 .. code-block:: console
449 * lsc_poll_period_ms: Optional parameter which defines the polling interval
450 in milli-seconds at which devices which don't support lsc interrupts are
451 checked for a change in the devices link status
453 .. code-block:: console
455 lsc_poll_period_ms=100
457 * up_delay: Optional parameter which adds a delay in milli-seconds to the
458 propagation of a devices link status changing to up, by default this
461 .. code-block:: console
465 * down_delay: Optional parameter which adds a delay in milli-seconds to the
466 propagation of a devices link status changing to down, by default this
469 .. code-block:: console
476 Create a bonded device in round robin mode with two slaves specified by their PCI address:
478 .. code-block:: console
480 $RTE_TARGET/app/testpmd -l 0-3 -n 4 --vdev 'net_bonding0,mode=0,slave=0000:0a:00.01,slave=0000:04:00.00' -- --port-topology=chained
482 Create a bonded device in round robin mode with two slaves specified by their PCI address and an overriding MAC address:
484 .. code-block:: console
486 $RTE_TARGET/app/testpmd -l 0-3 -n 4 --vdev 'net_bonding0,mode=0,slave=0000:0a:00.01,slave=0000:04:00.00,mac=00:1e:67:1d:fd:1d' -- --port-topology=chained
488 Create a bonded device in active backup mode with two slaves specified, and a primary slave specified by their PCI addresses:
490 .. code-block:: console
492 $RTE_TARGET/app/testpmd -l 0-3 -n 4 --vdev 'net_bonding0,mode=1,slave=0000:0a:00.01,slave=0000:04:00.00,primary=0000:0a:00.01' -- --port-topology=chained
494 Create a bonded device in balance mode with two slaves specified by their PCI addresses, and a transmission policy of layer 3 + 4 forwarding:
496 .. code-block:: console
498 $RTE_TARGET/app/testpmd -l 0-3 -n 4 --vdev 'net_bonding0,mode=2,slave=0000:0a:00.01,slave=0000:04:00.00,xmit_policy=l34' -- --port-topology=chained