EtherChannel and IEEE 802.3ad Link Aggregation are network port
aggregation technologies that allow several Ethernet adapters to be
aggregated together to form a single pseudo Ethernet device. For
example,
ent0 and
ent1 can be aggregated into an EtherChannel adapter called
ent3; interface
en3
would then be configured with an IP address. The system considers these
aggregated adapters as one adapter. Therefore, IP is configured over
them as over any Ethernet adapter. In addition, all adapters in the
EtherChannel or Link Aggregation are given the same hardware (MAC)
address, so they are treated by remote systems as if they were one
adapter. Both EtherChannel and IEEE 802.3ad Link Aggregation require
support in the switch so it is aware which switch ports should be
treated as one.
The main benefit of EtherChannel and IEEE 802.3ad Link Aggregation is
that they have the network bandwidth of all of their adapters in a
single network presence. If an adapter fails, network traffic is
automatically sent on the next available adapter without disruption to
existing user connections. The adapter is automatically returned to
service on the EtherChannel or Link Aggregation when it recovers.
There are some differences between EtherChannel and IEEE 802.3ad Link Aggregation. Consider the differences given in
Table 3 to determine which would be best for your situation.
Table 3. Differences between EtherChannel and IEEE 802.3ad Link Aggregation.
EtherChannel | IEEE 802.3ad |
Requires switch configuration | Little, if any, configuration of switch required to form aggregation. Some initial setup of the switch may be required. |
Supports different packet distribution modes | Supports only standard distribution mode |
Beginning with AIX 5L with 5200-03, Dynamic Adapter
Membership functionality is available. This functionality allows you to
add or remove adapters from an EtherChannel without having to disrupt
any user connections. For more details, see
Dynamic Adapter Membership.
Supported Adapters
EtherChannel and IEEE 802.3ad Link Aggregation are supported on the following Ethernet adapters:
- 10/100 Mbps Ethernet PCI Adapter
- Universal 4-Port 10/100 Ethernet Adapter
- 10/100 Mbps Ethernet PCI Adapter II
- 10/100/1000 Base-T Ethernet PCI Adapter
- Gigabit Ethernet-SX PCI Adapter
- 10/100/1000 Base-TX Ethernet PCI-X Adapter
- Gigabit Ethernet-SX PCI-X Adapter
- 2-port 10/100/1000 Base-TX Ethernet PCI-X Adapter
- 2-port Gigabit Ethernet-SX PCI-X Adapter
Only the basic EtherChannel functionality (operating exclusively in
"standard" or "round-robin" mode without a backup) is supported in the
following Ethernet adapters:
- PCI Ethernet BNC/RJ-45 Adapter
- PCI Ethernet AUI/RJ-45 Adapter
Unless the AIX Release Notes specify otherwise, support for new adapters will be provided as those adapters are released.
Note:
Mixing
adapters of different speeds in the same EtherChannel, even if one of
them is operating as the backup adapter, is not officially supported.
This does not mean that such configurations
will not work. The EtherChannel driver will make every reasonable
attempt to work even in a mixed-speed scenario.
For information on configuring and using EtherChannel, see
EtherChannel. For more information on configuring and using IEEE 802.3ad Link Aggregation, see
IEEE 802.3ad Link Aggregation. For information on the different AIX and switch configuration combinations and the results they will produce, see
Interoperability Scenarios.
EtherChannel
The adapters that belong to an
EtherChannel must be connected to the same EtherChannel-enabled switch.
This switch must be manually configured to treat the ports that belong
to the EtherChannel as an aggregated link. Note that your switch
documentation may refer to this capability as "link aggregation" or
"trunking."
Traffic is distributed across the adapters in either the standard way
(where the adapter over which the packets are sent is chosen depending
on an algorithm) or on a round-robin basis (where packets are sent
evenly across all adapters). Incoming traffic is distributed in
accordance to the switch configuration and is not controlled by the
EtherChannel operation mode.
In AIX, you can configure multiple EtherChannels per system, but it is
required that all the links in one EtherChannel are attached to a single
switch. Because the EtherChannel cannot be spread across two switches,
the entire EtherChannel is lost if the switch is unplugged or fails. To
solve this problem, a new backup option available in AIX 5.2 and later
keeps the service running when the main EtherChannel fails. The backup
and EtherChannel adapters should be attached to different network
switches, which must be inter-connected for this setup to work properly.
In the event that all of the adapters in the EtherChannel fail, the
backup adapter will be used to send and receive all traffic. When any
link in the EtherChannel is restored, the service is moved back to the
EtherChannel.
For example,
ent0 and
ent1 could be configured as the main EtherChannel adapters, and
ent2 as the backup adapter, creating an EtherChannel called
ent3. Ideally,
ent0 and
ent1 would be connected to the same EtherChannel-enabled switch, and
ent2 would be connected to a different switch. In this example, all traffic sent over
en3 (the EtherChannel's interface) would be sent over
ent0 or
ent1 by default (depending on the EtherChannel's packet distribution scheme), whereas
ent2 will be idle. If at any time both
ent0 and
ent1 fail, all traffic would be sent over the backup adapter,
ent2. When either
ent0 or
ent1 recover, they will once again be used for all traffic.
Network Interface Backup, a mode of operation available for EtherChannel
in AIX 4.3.3 and AIX 5.1, protects against a single point of Ethernet
network failure. No special hardware is required to use Network
Interface Backup, but the backup adapter should be connected a separate
switch for maximum reliability. In Network Interface Backup mode, only
one adapter at a time is actively used for network traffic. The
EtherChannel tests the currently-active adapter and, optionally, the
network path to a user-specified node. When a failure is detected, the
next adapter will be used for all traffic. Network Interface Backup
provides detection and failover with no disruption to user connections.
Network Interface Backup was originally implemented as a mode in the
EtherChannel SMIT menu. In AIX 5.2 and later, the backup adapter
provides the equivalent function, so the mode was eliminated from the
SMIT menu. To configure network interface backup in AIX 5.2 and later,
see
Configure Network Interface Backup.
Configuring EtherChannel
Follow these steps to configure an EtherChannel.
Considerations
- You can have up to eight primary Ethernet adapters and only one backup Ethernet adapter per EtherChannel.
- You can configure multiple EtherChannels on a single system, but
each EtherChannel constitutes an additional Ethernet interface. The no command option, ifsize,
may need to be increased to include not only the Ethernet interfaces
for each adapter, but also any EtherChannels that are configured. In AIX
5.2 and earlier, the default ifsize is eight. In AIX 5.2 and later, the default size is 256.
- You can use any supported Ethernet adapter in an EtherChannel (see Supported Adapters).
However, the Ethernet adapters must be connected to a switch that
supports EtherChannel. See the documentation that came with your switch
to determine if it supports EtherChannel (your switch documentation may
refer to this capability also as link aggregation or trunking).
- All adapters in the EtherChannel should be configured for the same speed (100 Mbps, for example) and should be full duplex.
- The adapters used in the EtherChannel cannot be accessed by the
system after the EtherChannel is configured. To modify any of their
attributes, such as media speed, transmit or receive queue sizes, and so
forth, you must do so before including them in the EtherChannel.
- The adapters that you plan to use for your EtherChannel must not
have an IP address configured on them before you start this procedure.
When configuring an EtherChannel with adapters that were previously
configured with an IP address, make sure that their interfaces are in
the detach state. The adapters to be added to the EtherChannel cannot have interfaces configured in the up
state in the Object Data Manager (ODM), which will happen if their IP
addresses were configured using SMIT. This may cause problems bringing
up the EtherChannel when the machine is rebooted because the underlying
interface is configured before the EtherChannel with the information
found in ODM. Therefore, when the EtherChannel is configured, it finds
that one of its adapters is already being used. To change this, before
creating the EtherChannel, type smit chinet, select each of the interfaces of the adapters to be included in the EtherChannel, and change its state value to detach.
This will ensure that when the machine is rebooted the EtherChannel can
be configured without errors. For more information about ODM, see Object Data Manager (ODM) in AIX 5L Version 5.2 General Programming Concepts: Writing and Debugging Programs.
- If you will be using 10/100 Ethernet adapters in the EtherChannel,
you may need to enable link polling on those adapters before you add
them to the EtherChannel. Type smit chgenet at the command line. Change the Enable Link Polling value to yes, and press Enter.
Note:
In
AIX 5L with 5200-03 and later, enabling the link polling mechanism is
not necessary. The link poller will be started automatically.
- If you plan to use jumbo frames, you may need to enable this feature
in every adapter before creating the EtherChannel and in the
EtherChannel itself. Type smitty chgenet at the command line. Change the Enable Jumbo Frames value to yes
and press Enter. Do this for every adapter for which you want to enable
Jumbo Frames. You will enable jumbo frames in the EtherChannel itself
later.
Note:
In
AIX 5.2 and later, enabling the jumbo frames in every underlying
adapter is not necessary once it is enabled in the EtherChannel itself.
The feature will be enabled automatically if you set the Enable Jumbo Frames attribute to yes.
Configure an EtherChannel
- Type smit etherchannel at the command line.
- Select Add an EtherChannel / Link Aggregation from the list and press Enter.
- Select the primary Ethernet adapters that you want on your
EtherChannel and press Enter. If you are planning to use EtherChannel
backup, do not select the adapter that you plan to use for the backup at
this point. The EtherChannel backup option is available in AIX 5.2 and
later.
Note:
The Available Network Adapters
displays all Ethernet adapters. If you select an Ethernet adapter that
is already being used (has an interface defined), you will get an error
message. You first need to detach this interface if you want to use it.
- Enter the information in the fields according to the following guidelines:
- EtherChannel / Link Aggregation Adapters: You should see all primary adapters that you are using in your EtherChannel. You selected these adapters in the previous step.
- Enable Alternate Address: This field is optional. Setting this to yes will enable you to specify a MAC address that you want the EtherChannel to use. If you set this option to no, the EtherChannel will use the MAC address of the first adapter.
- Alternate Address: If you set Enable Alternate Address to yes, specify the MAC address that you want to use here. The address you specify must start with 0x and be a 12-digit hexadecimal address (for example, 0x001122334455).
- Enable Gigabit Ethernet Jumbo Frames: This
field is optional. In order to use this, your switch must support jumbo
frames. This will only work with a Standard Ethernet (en) interface,
not an IEEE 802.3 (et) interface. Set this to yes if you want to enable it.
- Mode: You can choose from the following modes:
- standard:
In this mode the EtherChannel uses an algorithm to choose which adapter
it will send the packets out on. The algorithm consists of taking a
data value, dividing it by the number of adapters in the EtherChannel,
and using the remainder (using the modulus operator) to identify the
outgoing link. The Hash Mode value determines which data value is fed
into this algorithm (see the Hash Mode attribute for an explanation of
the different hash modes). For example, if the Hash Mode is standard,
it will use the packet's destination IP address. If this is 10.10.10.11
and there are 2 adapters in the EtherChannel, (1 / 2) = 0 with
remainder 1, so the second adapter is used (the adapters are numbered
starting from 0). The adapters are numbered in the order they are listed
in the SMIT menu. This is the default operation mode.
- round_robin: In this mode the EtherChannel
will rotate through the adapters, giving each adapter one packet before
repeating. The packets may be sent out in a slightly different order
than they were given to the EtherChannel, but it will make the best use
of its bandwidth. It is an invalid combination to select this mode with a
Hash Mode other than default. If you choose the round-robin mode, leave the Hash Mode value as default.
- netif_backup: This option is available
only in AIX 5.1 and AIX 4.3.3. In this mode, the EtherChannel will
activate only one adapter at a time. The intention is that the adapters
are plugged into different Ethernet switches, each of which is capable
of getting to any other machine on the subnet or network. When a problem
is detected either with the direct connection (or optionally through
the inability to ping a machine), the EtherChannel will deactivate the
current adapter and activate a backup adapter. This mode is the only one
that makes use of the Internet Address to Ping, Number of Retries, and Retry Timeout
fields. Network Interface Backup Mode does not exist as an explicit
mode in AIX 5.2 and later. To enable Network Interface Backup Mode in
AIX 5.2 and later, you must configure one adapter in the main
EtherChannel and a backup adapter. For more information, see Configure Network Interface Backup.
- 8023ad: This options enables the use of
the IEEE 802.3ad Link Aggregation Control Protocol (LACP) for automatic
link aggregation. For more details about this feature, see IEEE 802.3ad Link Aggregation.
- Hash Mode: You can choose from the
following hash modes, which will determine which data value will be used
by the algorithm to determine the outgoing adapter:
To learn more about packet distribution and load balancing, see Load-balancing options.
- Backup Adapter: This field is optional.
Enter the adapter that you want to use as your EtherChannel backup.
EtherChannel backup is available in AIX 5.2 and later.
- Internet Address to Ping: This field is optional and only takes effect if you are running Network Interface Backup
mode or if you have only one adapter in the EtherChannel and a backup
adapter. The EtherChannel will ping the IP address or host name that you
specify here. If the EtherChannel is unable to ping this address for
the Number of Retries times in Retry Timeout intervals, the EtherChannel will switch adapters.
- Number of Retries: Enter the number of
ping response failures that are allowed before the EtherChannel switches
adapters. The default is three. This field is optional and valid only
if you have set an Internet Address to Ping.
- Retry Timeout: Enter the number of seconds between the times when the EtherChannel will ping the Internet Address to Ping. The default is one second. This field is optional and valid only if you have set an Internet Address to Ping.
- Press Enter after changing the desired fields to create the EtherChannel.
- Configure IP over the newly-created EtherChannel device by typing smit chinet at the command line.
- Select your new EtherChannel interface from the list.
- Fill in all the required fields and press Enter.
Configure Network Interface Backup
Network
Interface Backup protects against a single point of network failure by
providing failure detection and failover with no disruption to user
connections. When operating in this mode, only one adapter is active at
any given time. If the active adapter fails, another adapter in the
EtherChannel will be used for all traffic. When operating in Network
Interface Backup mode, it is not necessary to connect to
EtherChannel-enabled switches.
The Network Interface Backup setup is most effective when the adapters
are connected to different network switches, as this provides greater
redundancy than connecting all adapters to one switch. When connecting
to different switches, make sure there is a connection between the
switches. This provides failover capabilities from one adapter to
another by ensuring that there is always a route to the currently-active
adapter.
In releases prior to AIX 5.2, Network Interface Backup mode was
implemented as an explicit mode of operation in the EtherChannel SMIT
menu. In AIX 5.2 and later, however, the backup adapter functionality
provides the equivalent behavior, so the mode was eliminated from the
SMIT menu.
Additionally, AIX 5.2 and later versions provide priority, meaning that
the adapter configured in the primary EtherChannel will be used
preferentially over the backup adapter. As long as the primary adapter
is functional, it will be used. This contrasts from the behavior of
Network Interface Backup mode in releases prior to AIX 5.2, where the
backup adapter was used until it also failed, regardless of whether the
primary adapter had already recovered.
For example,
ent0 could be configured as the main adapter, and
ent2 as the backup adapter, creating an EtherChannel called
ent3. Ideally,
ent0 and
ent2 would be connected to two different switches. In this example, all traffic sent over
en3 (the EtherChannel's interface) would be sent over
ent0 by default, whereas
ent2 will be idle. If at any time
ent0 fails, all traffic would be sent over the backup adapter,
ent2. When
ent0 recovers, it will once again be used for all traffic.
While operating in Network Interface Backup Mode, it is also possible to
configure the EtherChannel to detect link failure and network
unreachability. To do this, specify the IP address or host name of a
remote host where connectivity should always be present. The
EtherChannel will periodically ping this host to determine whether there
is still a network path to it. If a specified number of ping attempts
go unanswered, the EtherChannel will fail over to the other adapter in
the hope that there is a network path to the remote host through the
other adapter. In this setup, not only should every adapter be connected
to a different switch, but each switch should also have a different
route to the host that is pinged.
This ping feature is only available in Network Interface Backup mode.
However, in AIX 5.2 and later, if there is a failover due to unanswered
pings on the primary adapter, the backup adapter will remain the active
channel as long as it is working. There is no way of knowing, while
operating on the backup adapter, whether it is possible to reach the
host being pinged from the primary adapter. To avoid failing over back
and forth between the primary and the backup, it will simply keep
operating on the backup (unless the pings go unanswered on the backup
adapter as well, or if the backup adapter itself fails, in which case it
would fail over to the primary adapter). However, if the failover
occurred because the primary adapter failed (not because the pings went
unanswered), the EtherChannel will then come back to the primary adapter
as soon it has come back up, as usual.
To configure Network Interface Backup in AIX 5.2, see
Configure Network Interface Backup in AIX 5.2 and later. To configure Network Interface Backup in previous versions of AIX, see
Appendix B. Configure Network Interface Backup in previous AIX versions
Configure Network Interface Backup in AIX 5.2 and later
- With root authority, type smit etherchannel on the command line.
- Select Add an EtherChannel / Link Aggregation from the list and press Enter.
- Select the primary Ethernet adapter and press Enter. This is the adapter that will be used until it fails.
Note:
The
Available Network Adapters
displays all Ethernet adapters. If you select an Ethernet adapter that
is already being used, you will get an error message and will need to
detach this interface before you can use it. See the
ifconfig command for information on how to detach an interface.
- Enter the information in the fields according to the following guidelines:
- EtherChannel / Link Aggregation Adapters: You should see the primary adapter you selected in the previous step.
- Enable Alternate Address: This field is optional. Setting this to yes will enable you to specify a MAC address that you want the EtherChannel to use. If you set this option to no, the EtherChannel will use the MAC address of the primary adapter.
- Alternate Address: If you set Enable Alternate Address to yes, specify the MAC address that you want to use here. The address you specify must start with 0x and be a 12-digit hexadecimal address (for example 0x001122334455).
- Enable Gigabit Ethernet Jumbo Frames: This
field is optional. In order to use this, your switch must support jumbo
frames. This will only work with a Standard Ethernet (en) interface,
not an IEEE 802.3 (et) interface. Set this to yes if you want to use it.
- Mode: It is irrelevant which mode of
operation you select because there is only one adapter in the main
EtherChannel. All packets will be sent over that adapter until it fails.
There is no netif_backup mode because that mode can be emulated using a backup adapter.
- Hash Mode: It is irrelevant which hash
mode you select because there is only one adapter in the main
EtherChannel. All packets will be sent over that adapter until it fails.
- Backup Adapter: Enter the adapter that you
want to be your backup adapter. After a failover, this adapter will be
used until the primary adapter recovers. It is recommended to use the
preferred adapter as the primary adapter.
- Internet Address to Ping: The field is
optional. The EtherChannel will ping the IP address or host name that
you specify here. If the EtherChannel is unable to ping this address for
Number of Retries times in Retry Timeout intervals, the EtherChannel will switch adapters.
- Number of Retries: Enter the number of
ping response failures that are allowed before the EtherChannel switches
adapters. The default is three. This field is optional and valid only
if you have set an Internet Address to Ping.
- Retry Timeout: Enter the number of seconds between the times when the EtherChannel will ping the Internet Address to Ping. The default is one second. This field is optional and valid only if you have set an Internet Address to Ping.
- Press Enter after changing the desired fields to create the EtherChannel.
- Configure IP over the newly-created interface by typing smit chinet at the command line.
- Select your new EtherChannel interface from the list.
- Fill in all the required fields and press Enter.
For additional tasks that can be performed after the EtherChannel is configured, see
Managing EtherChannel and IEEE 802.3ad Link Aggregation.
Load-balancing options
There are two load
balancing methods for outgoing traffic in EtherChannel, as follows:
round-robin, which spreads the outgoing traffic evenly across all the
adapters in the EtherChannel; and standard, which selects the adapter
using an algorithm. The Hash Mode parameter determines which numerical
value is fed to the algorithm.
The following table summarizes the valid load balancing option combinations offered.
Table 4. Mode and Hash Mode combinations and the outgoing traffic distributions each will produce.
Mode | Hash Mode | Outgoing Traffic Distribution |
standard or 8023ad | default | The
traditional AIX behavior. The adapter selection algorithm uses the last
byte of the destination IP address (for TCP/IP traffic) or MAC address
(for ARP and other non-IP traffic). This mode is typically a good
initial choice for a server with a large number of clients. |
standard or 8023ad | src_dst_port | The
outgoing adapter path is selected by an algorithm using the combined
source and destination TCP or UDP port values. Since each connection has
a unique TCP or UDP port, the three port-based hash modes provide
additional adapter distribution flexibility when there are several,
separate TCP or UDP connections between an IP address pair. |
standard or 8023ad | src_port | The adapter selection algorithm uses the source TCP or UDP port value. In the netstat -an command output, the port is the TCP/IP address suffix value in the Local column. |
standard or 8023ad | dst_port | The outgoing adapter path is selected by the algorithm using the destination system port value. In the netstat -an command output, the TCP/IP address suffix in the Foreign column is the TCP or UDP destination port value. |
round-robin | default | Outgoing
traffic is spread evenly across all the adapter ports in the
EtherChannel. This mode is the typical choice for two hosts connected
back-to-back (without an intervening switch). |
Round-Robin
All outgoing traffic is spread evenly
across all of the adapters in the EtherChannel. It provides the highest
bandwidth optimization for the AIX server system. While round-robin
distribution is the ideal way to utilize all the links equally, consider
that it also introduces the potential for out-of-order packets at the
receiving system.
In general, round-robin mode is ideal for back-to-back connections
running jumbo frames. In this environment, there is no intervening
switch, so there is no chance that processing at the switch could alter
the packet delivery time, order, or adapter path. On this direct cable
network path, packets are received exactly as sent. Jumbo frames (9000
byte MTU) always yield better file transfer performance than traditional
1500 byte MTUs. In this case, however, they add another benefit. These
larger packets take longer to send so it is less likely that the
receiving host would be continuously interrupted with out-of-order
packets.
Round-robin mode can be implemented in other environments but at
increased risk of out-of-order packets at the receiving system. This
risk is particularly high when there are few, long-lived, streaming TCP
connections. When there are many such connections between a host pair,
packets from different connections could be intermingled, thereby
decreasing the chance of packets for the same connection arriving
out-of-order. Check for out-of-order packet statistics in the
tcp section of the
netstat -s command output. A steadily-increasing value indicates a potential problem in traffic sent from an EtherChannel.
If out-of-order packets are a problem on a system that must use
traditional Ethernet MTUs and must connected through a switch, try the
various hash modes offered in standard mode operation. Each mode has a
particular strength, but the default and src_dst_port modes are the
logical starting points as they are more widely applicable.
Standard or 8032ad
Standard algorithm.
The standard algorithm is used for both standard and IEEE 802.3ad-style
link aggregations. AIX divides the last byte of the "numerical value"
by the number of adapters in the EtherChannel and uses the remainder to
identify the outgoing link. If the remainder is zero, the first adapter
in the EtherChannel is selected; a remainder of one means the second
adapter is selected, and so on (the adapters are selected in the order
they are listed in the
adapter_names attribute).
The Hash Mode selection determines the numerical value used in the
calculation. By default, the last byte of the destination IP address or
MAC address is used in the calculation, but the source and destination
TCP or UDP port values may also be used. These alternatives allow you to
fine-tune the distribution of outgoing traffic across the real adapters
in the EtherChannel.
In default hash mode, the adapter selection algorithm is applied to the
last byte of the destination IP address for IP traffic. For ARP and
other non-IP traffic, the same formula is applied on the last byte of
the destination MAC address. Unless there is an adapter failure which
causes a failover, all traffic between a host pair in default standard
mode goes out over the same adapter. The default hash mode may be ideal
when the local host establishes connections to many different IP
addresses.
If the local host establishes lengthy connections to few IP addresses,
however, you will notice that some adapters carry a greater load than
others, because all the traffic sent to a specific destination is sent
over the same adapter. While this prevents packets from arriving
out-of-order, it may not utilize bandwidth in the most effective fashion
in all cases. The port-based hash modes still send packets in order,
but they allow packets belonging to different UDP or TCP connections,
even if they are sent to the same destination, to be sent over different
adapters, thus utilizing better the bandwidth of all the adapters.
In
src_dst_port hash mode, the TCP or UDP
source and destination port values of the outgoing packet are added,
then divided by two. The resultant whole number (no decimals) is plugged
into the standard algorithm. TCP or UDP traffic is sent on the adapter
selected by the standard algorithm and selected hash mode value. Non-TCP
or UDP traffic will fall back to the default hash mode, meaning the
last byte of either the destination IP address or MAC address. The
src_dst_port
hash mode option considers both the source and the destination TCP or
UDP port values. In this mode, all of the packets in one TCP or UDP
connection are sent over a single adapter so they are guaranteed to
arrive in order, but the traffic is still spread out because connections
(even to the same host) may be sent over different adapters. The
results of this hash mode are not skewed by the connection establishment
direction because it uses both the source and destination TCP or UDP
port values.
In
src_port hash mode, the source TCP or UDP port value of the outgoing packet is used. In
dst_port hash mode, the destination TCP or UDP port value of the outgoing packet is used. Use the
src_port or
dst_port hash mode options if port values change from one connection to another and if the
src_dst_port option is not yielding a desirable distribution.
Managing EtherChannel and IEEE 802.3ad Link Aggregation
This section will tell you how to perform the following tasks:
Listing EtherChannels or Link Aggregations
- On the command line, type smit etherchannel.
- Select List All EtherChannels / Link Aggregations and press Enter.
Changing the Alternate Address
This enables you to specify a MAC address for your EtherChannel or Link Aggregation.
- On AIX 5.2 with 5200-01 and earlier, type ifconfig interface detach, where interface
is your EtherChannel's or Link Aggregation's interface. (On AIX 5L with
5200-03 and later, you can change the alternate address of the
EtherChannel without detaching its interface).
- On the command line, type smit etherchannel.
- Select Change / Show Characteristics of an EtherChannel and press Enter.
- If you have multiple EtherChannels, select the EtherChannel for which you want to create an alternate address.
- Change the value in Enable Alternate EtherChannel Address to yes.
- Enter the alternate address in the Alternate EtherChannel Address field. The address must start with 0x and be a 12-digit hexadecimal address (for example, 0x001122334455).
- Press Enter to complete the process.
Note:
Changing
the EtherChannel's MAC address at runtime may cause a temporary loss of
connectivity. This is because the adapters need to be reset so they
learn of their new hardware address, and some adapters take a few
seconds to be initialized.
Dynamic Adapter Membership
Prior to AIX 5L
with 5200-03, in order to add or remove an adapter from an EtherChannel,
its interface first had to be detached, temporarily interrupting all
user traffic. To overcome this limitation, Dynamic Adapter Membership
(DAM) was added in AIX 5L with 5200-03. It allows adapters to be added
or removed from an EtherChannel without having to disrupt any user
connections. A backup adapter can also be added or removed; an
EtherChannel can be initially created without a backup adapter, and one
can be added a later date if the need arises
Not only can adapters be added or removed without disrupting user
connections, it is also possible to modify most of the EtherChannel
attributes at runtime. For example, you may begin using the "ping"
feature of Network Interface Backup while the EtherChannel is in use, or
change the remote host being pinged at any point.
You may also turn a regular EtherChannel into an IEEE 802.3ad Link
Aggregation (or vice versa), allowing users to experiment with this
feature without having to remove and recreate the EtherChannel.
Furthermore, with DAM, you may choose to create a one-adapter
EtherChannel. A one-adapter EtherChannel behaves exactly like a regular
adapter; however, should this adapter ever fail, it would be possible to
replace it at runtime without ever losing connectivity. To accomplish
this, you would add a temporary adapter to the EtherChannel, remove the
defective adapter from the EtherChannel, replace the defective adapter
with a working one using Hot Plug, add the new adapter to the
EtherChannel, and then remove the temporary adapter. During this process
you would never notice a loss in connectivity. If the adapter had been
working as a standalone adapter, however, it would have had to be
detached before being removed using Hot Plug, and during that time any
traffic going over it would simply have been lost.
Adding, removing, or changing adapters in an EtherChannel or Link Aggregation
There
are two ways to add, remove, or change an adapter in an EtherChannel or
Link Aggregation. One method requires the EtherChannel or Link
Aggregation interface to be detached, while the other does not (using
Dynamic Adapter Membership, which is available in AIX 5L with 5200-03
and later).
Making changes to an EtherChannel using Dynamic Adapter Membership
Making
changes using Dynamic Adapter Membership does not require you to stop
all traffic going over the EtherChannel by detaching its interface.
Consider the following before proceeding:
Notes:
- When
adding an adapter at runtime, note that different Ethernet adapters
support different capabilities (for example, the ability to do checksum
offload, to use private segments, to do large send, and so forth). If
different types of adapters are used in the same EtherChannel, the
capabilities reported to the interface layer are those supported by all
the adapters (for example, if all but one adapter supports the use of
private segments, the EtherChannel will state it does not support
private segments; if all adapters do support large send, the channel
will state it supports large send). When adding an adapter to an
EtherChannel at runtime, be sure that it supports at least the same
capabilities as the other adapters already in the EtherChannel. If you
attempt to add an adapter that does not support all the capabilities the
EtherChannel supports, the addition will fail. Note, however, that if
the EtherChannel's interface is detached, you may add any adapter
(regardless of which capabilities it supports), and when the interface
is reactivated the EtherChannel will recalculate which capabilities it
supports based on the new list of adapters.
- If you are not using an alternate address and you plan to delete the
adapter whose MAC address was used for the EtherChannel (the MAC
address used for the EtherChannel is "owned" by one of the adapters),
the EtherChannel will use the MAC address of the next adapter available
(in other words, the one that becomes the first adapter after the
deletion, or the backup adapter in case all main adapters are deleted).
For example, if an EtherChannel has main adapters ent0 and ent1 and backup adapter ent2, it will use by default ent0's MAC address (it is then said that ent0 "owns" the MAC address). If ent0 is deleted, the EtherChannel will then use ent1's MAC address. If ent1 is then deleted, the EtherChannel will use ent2's MAC address. If ent0 were later re-added to the EtherChannel, it will continue to use ent2's MAC address because ent2 is now the owner of the MAC address. If ent2 were then deleted from the EtherChannel, it would start using ent0's
MAC address again. Deleting the adapter whose MAC address was used for
the EtherChannel may cause a temporary loss of connectivity, because
all the adapters in the EtherChannel need to be reset so they learn of
their new hardware address. Some adapters take a few seconds to be
initialized.
If your EtherChannel is using an alternate address (a MAC address you
specified), it will keep using this MAC address regardless of which
adapters are added or deleted. Furthermore, it means that there will be
no temporary loss of connectivity when adding or deleting adapters
because none of the adapters "owns" the EtherChannel's MAC address.
- Almost all EtherChannel attributes can now be modified at runtime. The only exception is Enable Gigabit Ethernet Jumbo Frames. To modify the Enable Gigabit Ethernet Jumbo Frames attribute, you must first detach the EtherChannel's interface before attempting to modify this value.
- For any attribute that cannot be changed at runtime (currently, only Enable Gigabit Ethernet Jumbo Frames), there is a field called Apply change to DATABASE only. If this attribute is set to yes, it is possible to change, at runtime, the value of an attribute that usually cannot be modified at runtime. With the Apply change to DATABASE only field set to yes
the attribute will only be changed in the ODM and will not be reflected
in the running EtherChannel until it is reloaded into memory (by
detaching its interface, using rmdev -l EtherChannel_device and then mkdev -l EtherChannel_device
commands), or until the machine is rebooted. This is a convenient way
of making sure that the attribute is modified the next time the machine
boots, without having to disrupt the running EtherChannel.
To make changes to the EtherChannel or Link Aggregation using Dynamic Adapter Membership, follow these steps:
- At the command line, type smit etherchannel.
- Select Change / Show Characteristics of an EtherChannel / Link Aggregation.
- Select the EtherChannel or Link Aggregation that you want to modify.
- Fill in the required fields according to the following guidelines:
- In the Add adapter or Remove adapter field, select the Ethernet adapter you want to add or remove.
- In the Add backup adapter or Remove backup adapter fields, select the Ethernet adapter you want to start or stop using as a backup.
- Almost all the EtherChannel attributes may be modified at runtime, although the Enable Gigabit Ethernet Jumbo Frames attribute cannot.
- To turn a regular EtherChannel into an IEEE 802.3ad Link Aggregation, change the Mode attribute to 8023ad. To turn an IEEE 802.3ad Link Aggregation into an EtherChannel, change the Mode attribute to standard or round_robin.
- Fill in the necessary data, and press Enter.
Making changes on AIX 5.2 with 5200-01 and earlier
Follow these steps to detach the interface before making changes:
- Type ifconfig interface detach, where interface is your EtherChannel's interface.
- On the command line type, smit etherchannel.
- Select Change / Show Characteristics of an EtherChannel / Link Aggregation and press Enter.
- Select the EtherChannel or Link Aggregation that you want to modify.
- Modify the attributes you want to change in your EtherChannel or Link Aggregation and press Enter.
- Fill in the necessary fields and press Enter.
Remove an EtherChannel or Link Aggregation
- Type ifconfig interface detach, where interface is your EtherChannel's interface.
- On the command line type smit etherchannel.
- Select Remove an EtherChannel / and press Enter.
- Select the EtherChannel that you want to remove and press Enter.
Configure or remove a backup adapter on an existing EtherChannel or Link Aggregation
The
following procedure configures or removes a backup adapter on an
EtherChannel or Link Aggregation. This option is available only in AIX
5.2 and later.
- Type ifconfig interface detach, where interface is your EtherChannel's or Link Aggregation's interface.
- On the command line, type smit etherchannel.
- Select Change / Show Characteristics of an EtherChannel / Link Aggregation.
- Select the EtherChannel or Link Aggregation that you are adding or modifying the backup adapter on.
- Enter the adapter that you want to use as your backup adapter in the Backup Adapter field, or select NONE if you wish to stop using the backup adapter.
Troubleshooting EtherChannel
If you are having trouble with your EtherChannel, consider the following:
Tracing EtherChannel
Use
tcpdump and
iptrace
to troubleshoot the EtherChannel. The trace hook id for the
transmission packets is 2FA and for other events is 2FB. You cannot
trace receive packets on the EtherChannel as a whole, but you can trace
each adapter's receive trace hooks.
Viewing EtherChannel Statistics
Use the
entstat command to get the aggregate statistics of all the adapters in the EtherChannel. For example,
entstat ent3 will display the aggregate statistics of ent3. Adding the
-d flag will also display the statistics of each adapter individually. For example, typing
entstat -d ent3
will show you the aggregate statistics of the EtherChannel as well as
the statistics of each individual adapter in the EtherChannel.
Note:
In the General Statistics section, the number shown in Adapter Reset Count
is the number of failovers. In EtherChannel backup, coming back to the
main EtherChannel from the backup adapter is not counted as a failover.
Only failing over from the main channel to the backup is counted. In the
Number of Adapters field, the backup adapter is counted in the number displayed.
Improving Slow Failover
If the failover time when you
are using network interface backup mode or EtherChannel backup is slow,
verify that your switch is not running the Spanning Tree Protocol
(STP). When the switch detects a change in its mapping of switch port to
MAC address, it runs the spanning tree algorithm to see if there are
any loops in the network. Network Interface Backup and EtherChannel
backup may cause a change in the port to MAC address mapping.
Switch ports have a forwarding delay counter that determines how soon
after initialization each port should begin forwarding or sending
packets. For this reason, when the main channel is re-enabled, there is a
delay before the connection is re-established, whereas the failover to
the backup adapter is faster. Check the forwarding delay counter on your
switch and make it as small as possible so that coming back to the main
channel occurs as fast as possible.
For the EtherChannel backup function to work correctly, the forwarding
delay counter must not be more than 10 seconds, or coming back to the
main EtherChannel might not work correctly. Setting the forwarding delay
counter to the lowest value allowed by the switch is recommended.
Adapters not Failing Over
If adapter failures are not
triggering failovers and you are running AIX 5.2 with 5200-01 or
earlier, check to see if your adapter card needs to have link polling
enabled to detect link failure. Some adapters cannot automatically
detect their link status. To detect this condition, these adapters must
enable a link polling mechanism that starts a timer that periodically
verifies the status of the link. Link polling is disabled by default.
For EtherChannel to work correctly with these adapters, however, the
link polling mechanism must be enabled on each adapter before the
EtherChannel is created. If you are running AIX 5L with 5200-03 and
later, the link polling is started automatically and this cannot be an
issue.
Adapters that have a link polling mechanism have an ODM attribute called
poll_link, which must be set to
yes
for the link polling to be enabled. Before creating the EtherChannel,
use the following command on every adapter to be included in the
channel:
smit chgenet
Change the Enable Link Polling value to yes and press Enter.
Using Jumbo Frames
For the jumbo frames option to work properly in AIX 5.2 and earlier, aside from enabling the
use_jumbo_frame
attribute on the EtherChannel, you must also enable jumbo frames on
each adapter before creating the EtherChannel using the following
command:
smitty chgenet
Change the Enable Jumbo Frames value to yes
and press Enter. On AIX 5.2 and later, jumbo frames are enabled
automatically in every underlying adapter when it is set to yes.
Remote Dump
Remote dump is not supported over an EtherChannel.
IEEE 802.3ad Link Aggregation
IEEE 802.3ad is a
standard way of doing link aggregation. Conceptually, it works the same
as EtherChannel in that several Ethernet adapters are aggregated into a
single virtual adapter, providing greater bandwidth and protection
against failures. For example,
ent0 and
ent1 can be aggregated into an IEEE 802.3ad Link Aggregation called
ent3; interface
en3
would then be configured with an IP address. The system considers these
aggregated adapters as one adapter. Therefore, IP is configured over
them as over any Ethernet adapter.
Like EtherChannel, IEEE 802.3ad requires support in the switch. Unlike
EtherChannel, however, the switch does not need to be configured
manually to know which ports belong to the same aggregation.
The advantages of using IEEE 802.3ad Link Aggregation instead of
EtherChannel are that it creates the link aggregations in the switch
automatically, and that it allows you to use switches that support the
IEEE 802.3ad standard but do not support EtherChannel.
In IEEE 802.3ad, the Link Aggregation Control Protocol (LACP)
automatically tells the switch which ports should be aggregated. When an
IEEE 802.3ad aggregation is configured, Link Aggregation Control
Protocol Data Units (LACPDUs) are exchanged between the server machine
and the switch. LACP will let the switch know that the adapters
configured in the aggregation should be considered as one on the switch
without further user intervention.
Although the IEEE 802.3ad specification does not allow the user to
choose which adapters are aggregated, the AIX implementation does allow
the user to select the adapters. According to the specification, the
LACP determines, completely on its own, which adapters should be
aggregated together (by making link aggregations of all adapters with
similar link speeds and duplexity settings). This prevents you from
deciding which adapters should be used standalone and which ones should
be aggregated together. The AIX implementation gives you control over
how the adapters are used, and it never creates link aggregations
arbitrarily.
To be able to aggregate adapters (meaning that the switch will allow
them to belong to the same aggregation) they must be of the same line
speed (for example, all 100 Mbps, or all 1 Gbps) and they must all be
full duplex. If you attempt to place adapters of different line speeds
or different duplex modes, the creation of the aggregation on the AIX
system will succeed, but the switch may not aggregate the adapters
together. If the switch does not successfully aggregate the adapters
together, you may notice a decrease in network performance. For
information on how to determine whether an aggregation on a switch has
succeeded, see
Troubleshooting IEEE 802.3ad.
According to the IEEE 802.3ad specification, packets going to the same
IP address are all sent over the same adapter. Thus, when operating in
8023ad mode, the packets will always be distributed in the standard fashion, never in a round-robin fashion.
The backup adapter feature is available for IEEE 802.3ad Link
Aggregations just as it is for EtherChannel. The backup adapter does not
need to be connected to an IEEE 802.3ad-enabled switch, but if it is,
the backup adapter will still follow the IEEE 802.3ad LACP.
You can also configure an IEEE 802.3ad Link Aggregation if the switch
supports EtherChannel but not IEEE 802.3ad. In that case, you would have
to manually configure the ports as an EtherChannel on the switch (just
as if a regular EtherChannel had been created). By setting the mode to
8023ad,
the aggregation will work with EtherChannel-enabled as well as IEEE
802.3ad-enabled switches. For more information about interoperability,
see
Interoperability Scenarios.
Note:
The
steps to enable the use of IEEE 802.3ad varies from switch to switch.
You should consult the documentation for your switch to determine what
initial steps, if any, must be performed to enable LACP in the switch.
For information in how to configure an IEEE 802.3ad aggregation, see
Configuring IEEE 802.3ad Link Aggregation.
Considerations
Consider the following before configuring an IEEE 802.3ad Link Aggregation:
Configuring IEEE 802.3ad Link Aggregation
Follow these steps to configure an IEEE 802.3ad Link Aggregation:
- Type smit etherchannel at the command line.
- Select Add an EtherChannel / Link Aggregation from the list and press Enter.
- Select the primary Ethernet adapters that you want on your Link
Aggregation and press Enter. If you are planning to use a backup
adapter, do not select the adapter that you plan to use for the backup
at this point. The backup adapter option is available in AIX 5.2 and
later.
Note:
The Available Network Adapters
displays all Ethernet adapters. If you select an Ethernet adapter that
is already being used (has an interface defined), you will get an error
message. You first need to detach these interfaces if you want to use
them.
- Enter the information in the fields according to the following guidelines:
- EtherChannel / Link Aggregation Adapters:
You should see all primary adapters that you are using in your Link
Aggregation. You selected these adapters in the previous step.
- Enable Alternate Address: This field is optional. Setting this to yes will enable you to specify a MAC address that you want the Link Aggregation to use. If you set this option to no, the Link Aggregation will use the MAC address of the first adapter.
- Alternate Address: If you set Enable Alternate Address to yes, specify the MAC address that you want to use here. The address you specify must start with 0x and be a 12-digit hexadecimal address (for example, 0x001122334455).
- Enable Gigabit Ethernet Jumbo Frames: This
field is optional. In order to use this, your switch must support jumbo
frames. This will only work with a Standard Ethernet (en) interface,
not an IEEE 802.3 (et) interface. Set this to yes if you want to enable it.
- Mode: Enter 8023ad.
- Hash Mode: You can choose from the
following hash modes, which will determine which data value will be used
by the algorithm to determine the outgoing adapter:
- default:
In this hash mode the destination IP address of the packet will be used
to determine the outgoing adapter. For non-IP traffic (such as ARP),
the last byte of the destination MAC address is used to do the
calculation. This mode will guarantee packets are sent out over the
EtherChannel in the order they were received, but it may not make full
use of the bandwidth.
- src_port: In this hash mode the source UDP
or TCP port value of the packet will be used to determine the outgoing
adapter. If the packet is not UDP or TCP traffic, the last byte of the
destination IP address will be used. If the packet is not IP traffic,
the last byte of the destination MAC address will be used.
- dst_port: In this hash mode the
destination UDP or TCP port value of the packet will be used to
determine the outgoing adapter. If the packet is not UDP or TCP traffic,
the last byte of the destination IP will be used. If the packet is not
IP traffic, the last byte of the destination MAC address will be used.
- src_dst_port: In this hash mode both the
source and destination UDP or TCP port values of the packet will be used
to determine the outgoing adapter (specifically, the source and
destination ports are added and then divided by two before being fed
into the algorithm). If the packet is not UDP or TCP traffic, the last
byte of the destination IP will be used. If the packet is not IP
traffic, the last byte of the destination MAC address will be used. This
mode can give good packet distribution in most situations, both for
clients and servers.
To learn more about packet distribution and load balancing, see Load-balancing options.
- Backup Adapter: This field is optional.
Enter the adapter that you want to use as your backup. The backup
adapter option is available in AIX 5.2 and later.
- Internet Address to Ping: This field is
optional, and only available if you have only one adapter in the main
aggregation and a backup adapter. The Link Aggregation will ping the IP
address or host name that you specify here. If the Link Aggregation is
unable to ping this address for the Number of Retries times in Retry Timeout intervals, the Link Aggregation will switch adapters.
- Number of Retries: Enter the number of
ping response failures that are allowed before the Link Aggregation
switches adapters. The default is three. This field is optional and
valid only if you have set an Internet Address to Ping.
- Retry Timeout: Enter the number of seconds between the times when the Link Aggregation will ping the Internet Address to Ping. The default is one second. This field is optional and valid only if you have set an Internet Address to Ping.
- Press Enter after changing the desired fields to create the Link Aggregation.
- Configure IP over the newly-created Link Aggregation device by typing smit chinet at the command line.
- Select your new Link Aggregation interface from the list.
- Fill in all the required fields and press Enter.
Managing IEEE 802.3ad
For management tasks that can be performed on an IEEE 802.3ad Link Aggregation after configuration, see
Managing EtherChannel and IEEE 802.3ad Link Aggregation.
Troubleshooting IEEE 802.3ad
If you are
having trouble with your IEEE 802.3ad Link Aggregation, use the
following command to verify the mode of operation of the Link
Aggregation:
entstat -d device
where device is the Link Aggregation device.
This
will also make a best-effort determination of the status of the
progress of LACP based on the LACPDUs received from the switch. The
following status values are possible:
- Inactive: LACP has not been initiated. This
is the status when a Link Aggregation has not yet been configured,
either because it has not yet been assigned an IP address or because its
interface has been detached.
- Negotiating: LACP is in progress, but the
switch has not yet aggregated the adapters. If the Link Aggregation
remains on this status for longer than one minute, verify that the
switch is correctly configured. For instance, you should verify that
LACP is enabled on the ports.
- Aggregated: LACP has succeeded and the switch has aggregated the adapters together.
- Failed: LACP has failed. Some possible causes
are that the adapters in the aggregation are set to different line
speeds or duplex modes or that they are plugged into different switches.
Verify the adapters' configuration. In addition, some switches allow
only contiguous ports to be aggregated and may have a limitation on the
number of adapters that can be aggregated. Consult the switch
documentation to determine any limitations that the switch may have,
then verify the switch configuration.
Note:
The
Link Aggregation status is a diagnostic value and does not affect the
AIX side of the configuration. This status value was derived using a
best-effort attempt. To debug any aggregation problems, it is best to
verify the switch's configuration.
Interoperability Scenarios
The following
table shows several interoperability scenarios. Consider these scenarios
when configuring your EtherChannel or IEEE 802.3ad Link Aggregation.
Additional explanation of each scenario is given after the table.
Table 5. Different AIX and switch configuration combinations and the results each combination will produce.
EtherChannel mode | Switch configuration | Result |
8023ad | IEEE 802.3ad LACP | OK - AIX initiates LACPDUs, which triggers an IEEE 802.3ad Link Aggregation on the switch. |
standard or round_robin | EtherChannel | OK - Results in traditional EtherChannel behavior. |
8023ad | EtherChannel | OK - Results in traditional EtherChannel behavior. AIX initiates LACPDUs, but the switch ignores them. |
standard or round_robin | IEEE 802.3ad LACP | Undesirable
- Switch cannot aggregate. The result may be poor performance as the
switch moves the MAC address between switch ports |
For more information visit :-
http://publib.boulder.ibm.com/infocenter/pseries/v5r3/index.jsp?topic=/com.ibm.aix.commadmn/doc/commadmndita/etherchannel_intro.htm