Network Parameters Boot Configuration
Structure of the ANS Build Tree
This file describes the Intel(R) Advanced Network Services (ANS) package, version 3.0.x, for Linux* systems. Intel ANS provides both teaming (including Multi-Vendor Teaming), and IEEE VLAN capabilities.
This file describes the Intel® Advanced Network Services (ANS) package, version 3.0.x, for Linux* systems. Intel ANS provides both teaming (including Multi-Vendor Teaming), and IEEE VLAN capabilities.
For ANS support, this package version must be used with the Linux base driver version e100-2.3.x for PRO/100 adapters and/or driver version e1000-5.2.x for PRO/1000 adapters. Base drivers provided in some distributions do not contain the MII interface that is necessary for ANS.
This package is intended for 2.4.7 and higher kernels; it was tested primarily on the Red Hat* Enterprise Linux ES 2.1 kernel 2.4.9-e.27 andThis version of ANS may be configured with the PROCfg command-line utility. This reporting and configuration tool reduces the command-line input required with the manual configuration method described in this document. At this time, PROCfg is only available for 32-bit architectures. For more information on using the PROCfg utility refer to procfg.htm. PROCfg is available on the Intel Customer Support website, http://support.intel.com.
|
NOTE: ANS was not tested with drivers that are statically linked into the kernels; > it was only tested with loadable modules. |
This package supports the following teaming modes:
Adapter Fault Tolerance (AFT). Default mode when a team is created and the mode is not specified. A secondary adapter in the team automatically takes over on any failure of the primary connection (cable, adapter or port). Only one network adapter is active at a time. After a failover, if the connection to the former primary adapter is restored, and it was set to be the preferred primary, control passes automatically back to that primary adapter, otherwise no failback is performed. May mix speed and duplex capabilities and settings. May use a hub or switch.
Switch Fault Tolerance (SFT). Provides a fail-over between switches when each adapter in the team is connected to a separate switch. Spanning Tree Protocol (STP) must be turned on in the switches and the forward delay parameter of iANS must be defined according to the switch settings. We recommend that you set the ANS forward delay parameter to twice the value of the STP forward delay configured on the switch. SFT teaming mode has been tested with the following configuration only: exactly two members in a team connected to two switches, with one member set as preferred primary and the other as preferred secondary; the switches’ setup was similar to that in the following diagram:
Adaptive Load Balancing (ALB). Includes Receive Load Balancing (RLB). A team of 2-8 adapters that performs receive and transmit load balancing with no required switch configuration. Supports mixed-speed/duplex settings among team members. Includes fault tolerance. An ALB team may be created with or without RLB enabled.
If RLB is enabled (default), all active adapters with the highest speeds load balance the IPv4 received traffic, and the primary adapter receives traffic using any protocol. Receive Load balancing is achieved by using the Address Resolution Protocol (ARP). Therefore, all packets coming from clients behind a router will be assigned the same receiving member that the router was assigned. For ALB teams with RLB disabled, only the primary adapter receives.
Intel Link Aggregation, Cisco*'s Fast EtherChannel* Technology or static 802.3ad (FEC or FEC/LA/802.3ad: static). A team of 2-8 10/100 adapters which simultaneously receive and transmit data. Includes fault tolerance and load balancing. Requires a switch that supports Intel Link Aggregation, Cisco*'s FEC or static 802.3ad. Must match speed/duplex settings on all team members. Must match switch aggregation requirements.
Gigabit equivalent of FEC (GEC or GEC/LA/802.3ad: static). Requires PRO/1000 or equivalent adapters and a switch that supports Intel Link Aggregation, Cisco*'s GEC or static 802.3ad. Other requirements are similar to FEC.
IEEE 802.3ad: dynamic (802.3ad). This mode is the IEEE standard for the technology incorporated into Cisco's FEC method. ANS support for dynamic 802.3ad teaming is similar to that for FEC and GEC teaming. In addition to the benefits of Link Aggregation, IEEE 802.3ad Dynamic Link Aggregation generally offers the following advantages: Automatic Configuration, Rapid Configuration and Reconfiguration, and Deterministic Behavior. It consists of a team of 2-8 adapters that simultaneously receive and transmit data. Aggregation groups include only members with full duplex and the same speed. Includes fault tolerance and load balancing. Requires a switch that supports the IEEE 802.3ad dynamic standard. Must match switch aggregation requirements.
NONE. A team of one Intel adapter that adds VLAN support on top of one existing interface. Does not require an Intel server adapter. VLAN-enabled must be set.
|
NOTES: You must have at least one Intel PRO/100 or
PRO/1000 server adapter to form a team. For AFT, SFT, and ALB, you may specify which adapters in a team you want to use as the primary and secondary adapters. The primary one is the main adapter used in a team. For AFT/SFT, the primary one is the only adapter used until a failure occurs. For ALB, the primary one is the adapter that both transmits and receives. For ALB+RLB, the primary adapter is the only adapter that receives non-IPv4 traffic. In a team of more than two adapters, the secondary adapter fills the primary's role if the primary adapter fails. Also, fail back will not occur unless the primary has been specified. If the priority is not specified before the team is activated, the adapter with the highest supported speed is chosen. If all adapters support the same speed, the last one added is the primary adapter. |
ANS supplies a periodic probes mechanism for checking the health of the system and discovering connectivity problems. The probes mechanism can help detect situations of wrong switch configuration: e.g., a port is disabled or has a wrong VLAN setting that otherwise would not be detected by ANS. Without probes enabled, ANS can only detect link loss and hardware/software failures reported by the base driver. You can use the probes mechanism for confirming your system's connectivity after configuration changes. Probes can be used in all non-Link Aggregation teaming modes, including AFT, SFT, and ALB(+RLB). Fine tuning of the probe parameters can scale failover time and enhance the system's reliability. Probes are on by default in teams with three or more adapters, and off by default otherwise. The team must include at least two adapters for the probes mechanism to work. If you create a team of two members and later add a third member, probes will not be enabled by default. See the PROCfg or ianscfg man page for information on how to change probes parameters.
|
NOTES: The probes mechanism can only indicate that there is connectivity between the members of the team. They do not indicate that there is connectivity to other clients on the network. When probes are enabled and the forward delay parameter of ANS is set, the status of the probes will override the forward delay and a member will be activated as soon as it transmits and receives probes successfully even if the forward delay is not up yet. In systems where connectivity between members does not indicate that the switch is ready to forward traffic to other clients (if forward delay could vary for different ports on the switch), probes must be turned off for the forward delay to be affective. |
Multi-Vendor Teaming (MVT) adds the capability to include non-Intel adapters from selected vendors in a team (AFT, SFT, ALB(+RLB), FEC, GEC, 802.3ad). MVT cannot be used with VLANs.
In order to activate MVT, you must have at least one Intel server adapter in the team that is designated as the primary adapter. Adapters from other vendors must be properly loaded. Use the same command line parameters for building the team as with all Intel adapters. All members in an MVT operate on a common feature set.
Base driver is required to support Ethtool version 1.2.
For high availability (link loss detection and current speed/duplex), the base
driver is also required to include support for one of the following options:
ANS defaults to the netif_carrier option if ethtool and MII support was not found.
|
NOTES: For RLB teams, some adapters produced by non-Intel vendors do not support setting the MAC address of the device while it is open. Therefore, swapping MAC addresses during failover is impossible. This creates a problem when RLB is enabled, as each team member must use a unique MAC address. As a workaround, ANS sets the non-Intel adapter to work in promiscuous mode when it becomes the primary after a failover occurs. The adapter will stay promiscuous as long as the MAC address that it is using differs from that of the adapter's hardware. |
| Using ALB and RLB with non-Intel adapters requires the use of a switch instead of a hub. | |
| Adding to an ANS team an adapter controlled by the tg3 driver
in Red Hat 2.1 has been found to cause a system panic due to
incompatibility. This problem was not found with the tg3 driver in Red Hat 3
or with the tg3 driver in the 2.4.23 vanilla kernel. Removing from an ANS team an active Broadcom NetExtreme 5700 or 5701 Gigabit adapter was found to cause reboot of the system on some occasions due to incompatibility. |
IEEE 802.3ac Virtual LANs (VLANs) are based on the addition of a tag to the packet header. All equipment on the network that passes this traffic must be capable of handling the extra length. Settings on the adapter must match the VLAN settings on the switch. There is a limit of 64 VLANs per server; however, VLAN IDs may be numbered from 1 to 4094.
|
NOTE: When using ANS VLANs, the native VLAN module provided with the kernel (8021q.o) must NOT be loaded. |
| VLAN hardware off-loading is supported only for kernels that support VLAN hwaccel. These include stock Red Hat 7.3 and later distribution kernels, or any 2.4.19 or later kernel. For older kernels, ANS implements software VLAN tagging and stripping. For VLAN hardware off-loading (enhanced performance) on older kernels you must use ANS driver version 1.2.x to 2.0.x, plus the appropriate base driver version(s) for that ANS driver. | |
| Due to a third-party limitation, adding VLANs when the IPv6 module is loaded might cause a network connectivity failure. |
Setting VLANs statically:
When working with VLAN tagging, all the ports on the switch connected to the server
should be configured to be tagged. In some Cisco
switches it is possible to configure ports to be both tagged and untagged at the
same time; although, this configuration will result in communication loss.
When adding new VLAN IDs to an active team the switch ports attached to the team's members must be configured with the new VLAN IDs prior to adding them to the ANS team. Otherwise, ANS might disable members that are not sensing probes on the new VLAN IDs because the switch is not configured accordingly.
Setting VLANs dynamically - GVRP:
VLANs may be set dynamically with the GARP VLAN Registration Protocol (GVRP).
GVRP is an industry-standard protocol designed to dynamically create and manage
VLAN information from device to device. With GVRP, a single switch is manually
configured with all the desired VLANs for the network, and all other switches on
the network obtain the VLAN information dynamically.
If GVRP is enabled in the team and in the switch, the switch ports connected to the server should not be marked as tagged, but will be dynamically configured by ANS whenever VLAN IDs are added and removed from the team.
Installation assumes all necessary build tools, kernel sources, and headers are in place and properly configured.
To compile and install the base driver package(s), see the readme included with those drivers. Ensure no other base drivers are loaded or compiled into the kernel for any of the Intel adapters (e.g., eepro, eepro100, eexpress, e100, e1000). Ensure the configuration files have been edited to alias the new driver(s). See the Example Setup section for an example file.
|
NOTE: Not for use with eepro100 (module removal may be necessary). |
Verify that all interfaces are working and that you have the base driver file(s), e100.o and/or e1000.o, located in:
/lib/modules/<kernel_version>/kernel/drivers/net/
The locations listed above may vary for certain Linux distributions. For exceptions, refer to ldistrib.htm.
To uninstall ANS, see Removing ANS.
Login as root.
Copy the iANS-x.x.x.tar.gz package file to a directory of your choice. "x.x.x" specifies the ANS version number.
Open the package file by running 'tar xzf iANS-x.x.x.tar.gz'. This creates a subdirectory for iANS-x.x.x under the current directory.
Enter the iANS-x.x.x/src directory. Run 'make' to compile the module.
|
NOTE: SMP/UP and MODVERSIONS are determined by 'make'. If there is a mismatch between the configuration for the kernel source tree and the running kernel, a warning is displayed. Run 'make options' to list the options available for overriding the source tree settings. Options include SMP, MODVERSIONS, and DEBUG (not listed). DEBUG values are 0 (debugging is off - default) and 1 (debugging is on). |
Run 'make install'.
|
NOTE: The default file locations are determined by 'make'. If there is a mismatch between the configuration for the kernel source tree and the running kernel, a warning is displayed. Either reboot to a kernel that matches the kernel source tree or set the symbolic link, usr/src/linux, to point to the kernel source tree that matches the running kernel. |
For a list of files and their default locations, see the Structure of the ANS Build Tree section in this file.
Follow these instructions to remove ANS:
Login as root.
Run 'ifconfig' to see every open virtual adapter configured in the system.
For each open virtual adapter, run 'ifconfig <vadapter_name> down'.
Remove the ANS module by running 'rmmod ians'.
Remove or disable each virtual adapter’s network boot file, as described in the Network Parameters Boot Configuration, Configuring ANS section.
From the iANS-x.x.x/src directory, run 'make uninstall' to remove all the files.
You can choose one of these three methods for configuring ANS:
PROCfg Utility. Provides a simplified configuration tool that reduces the amount of command-line input.
Scripted. Suggested for new users.
Manual, using the ianscfg utility.
ANS configuration and topology may be changed, with a few limitations. After a team is activated, you may change member priorities, add or remove team members, add or remove virtual adapters, or remove a team entirely. You may not:
Change the team mode.
Enable or disable RLB teaming mode.
Remove or add VLAN support.
Enable or disable GVRP.
Remove a team's last virtual adapter.
Remove the last member of a team with open virtual adapters.
Remove the last server adapter of a team with open virtual adapters.
If a team is left with no server adapter or with no members at all, it goes into an idle state until members are added. During such an idle period, resources used by the team are not freed.
For instructions on modifying ANS, see the instructions in either the scripted or manual configuration methods.
A member can be added or removed from a team when it is up and active with certain limitations:
Hot Add:
Hot Remove:
For ease of configuration, Intel supplies an interactive script to help configure an ANS topology. The script offers only the basic and most common configuration capabilities. To be able to use all available configuration options, refer to the Manual Configuration section, and to the ianscfg man page.
Before using the interactive script to create and modify an ANS topology, ensure the necessary base drivers (e.g. e100, e1000) are loaded.
To use the script, enter:
ianstool
You will be prompted with an interactive menu, offering several possible actions to be performed on the topology. These actions include:
Adding a team
Deleting an existing team
Viewing an existing team’s status
Configuring an existing team’s topology (i.e. adding/removing members and virtual adapters)
Saving the current topology
The script will prompt you for necessary information, such as the teaming mode, as you go along. As noted above, the script does not offer all possible configuration options that are available in ANS. For information on configuring options that are not available with the script, refer to the Manual Configuration section.
After the required base drivers are loaded, bring down any Ethernet adapters that are to be added to a team.
Command Description ifconfig -a
lists all current network devices ifconfig ethx down
brings down adapters so ANS may use them
|
|
NOTE: Do not change the configuration of Ethernet adapters that are not added to teams. This ensures that the network device name assignment remains consistent after rebooting. |
Load the ANS module:
insmod ians
Use the ianscfg utility to create and modify teams and VLANs. Refer to the ianscfg man page for specific usage details.
Activate your teams using ianscfg, if you have not done so already. Activate your virtual adapter(s), using:
ifconfig <vadapter_name> <IPaddr> [netmask <NETMASK>] [broadcast <BROADCAST>]
At this point you should be able to send and receive packets via the virtual adapter(s).
You can save the ANS topology or restore a previously saved topology using ianscfg. See the man page for details.
If you want the new configurations to take affect after each reboot, follow the instructions in Saving ANS Configuration Across Reboots.
You can also save
an ANS configuration across reboots from a previously
ianscfg –b [-f <file name>]
Using this
command will set the system to load the ANS configuration in the
|
|
NOTES: Saving the configuration across reboots either creates or modifies the network configuration files for the participating adapters. For SuSE* 8.0 (and up) distributions, these files’ locations and formats are assumed to be as specified in the Network Parameters Boot Configuration section, under _SuSE*_. For other distributions, the location and format are assumed to be as specified in the same section, under _RedHat*_. If your distribution uses a different location/format for network configuration files (e.g. SuSE* 7.x), ianscfg’ most likely will not be able to save the configuration across reboots. |
| If you modify the configuration such that some of the virtual adapters that previously existed no longer exist, the system will still make an (unsuccessful) attempt to load those (now non-existent) virtual adapters when you next boot the machine. To avoid this behavior, either use your distribution’s network configuration tools (e.g. netconf, setup) to disable (or remove) those virtual adapters, or simply remove the relevant network configuration files as specified in the Network Parameters Boot Configuration section. | |
| When saving your configuration across reboots, ANS creates a configuration file called '/etc/ians/ians-modules.conf'. You can safely remove this file if you no longer wish your ANS configuration to be automatically loaded after each reboot. |
If you want your virtual adapters' network parameters (i.e. IP address, netmask, etc.) to load after the next reboot, you must first ensure you have saved your topology to load on the next reboot. For details, see the Scripted Configuration and Manual Configuration sections.
You can save your virtual adapters' network parameters for boot in one of two ways:
Using the network and system configuration tools that are provided by your distribution (e.g. netconf, setup, etc.)
Manually edit or create the virtual adapters' boot configuration files (usually named ifcfg-<vadapter name>). This can be slightly different for different distributions. Following are specific examples for common distributions. If you have a different Linux distribution, the configuration files' location and syntax should still be similar (if nor identical) to one of the examples shown below. You should find the directory that includes the adapters' configuration files in your distribution, and follow the syntax of the other configuration files there.
Red Hat:
The files reside in /etc/sysconfig/network-scripts/.
Sample file content (for file ifcfg-<vadapter_name>):
DEVICE="<vadapter_name>"
BOOTPROTO="none"
ONBOOT="yes"
IPADDR="<IP_address>"
NETMASK="<netmask>"
SuSE:
The files reside in /etc/sysconfig/network/.
Sample file content (for file ifcfg-<vadapter_name>):
STARTMODE="onboot"
IPADDR="<IP_address>"
NETMASK="<netmask>"
Kernel dependant source files (src/*.[ch], src/lib/*.[ch])
Kernel independent precompiled object file (bin/[ARCH]/ians_core.o)
Makefiles
ANS configuration utility (bin/[ARCH]/ianscfg)
Configuration scripts (bin/iansonboot, bin/ianstool)
|
NOTE: [ARCH] is either "ia32" or "ia64". |
By default, the components are distributed as follows:
/lib/modules/<kernel_version>/kernel/drivers/net/ians.o
/usr/sbin/ianscfg
/usr/sbin/ianstool
/usr/sbin/iansonboot
/usr/share/man/man1/ianscfg.1.gz
#alias eth0 eepro100
#alias eth1 eepro100
#alias eth2 eepro100
alias eth0 e100
alias eth1 e100
alias eth2 e100
alias parport_lowlevel parport_pc
alias scsi_hostadapter aic7xxx
insmod e100
insmod ians
ianscfg -a -t team1 -M AFT -V
ianscfg -at team1 -m eth0 -p primary
ianscfg -at team1 -m eth1 -p secondary
ianscfg -at team1 -v vadapt1 -i 10
ianscfg -at team1 -v vadapt2 -i 15
ianscfg -c team1
ianscfg -s
ifconfig vadapt1 192.168.1.1 netmask 255.255.255.0
ifconfig vadapt2 192.168.2.1 netmask 255.255.255.0
Due to the default ARP behavior on Linux, it is not possible to have one
system on two IP networks in the same Ethernet broadcast domain
(non-partitioned switch) behave as expected. All Ethernet interfaces will
respond to IP traffic for any IP address assigned to the system.
This results in unbalanced receive traffic.
If you have multiple interfaces in a server, either turn on ARP filtering by
entering:
echo 1 > /proc/sys/net/ipv4/conf/all/arp_filter
(this only works if your kernel's version is higher than 2.4.5),
or,
install the interfaces in separate broadcast domains (either in different
switches or in a switch partitioned to VLANs).
"service network restart" will fail to configure a virtual adapter IP with DHCP if the virtual adapter interface name is lexicographically smaller than the interface names of team members (physical adapters). The system's network service runs the network scripts according to their lexicographical order, so the lexicographical order of the interface names determines whether the virtual adapters or members are brought up first. For successful DHCP configuration of a virtual adapter IP, the members interfaces must be up and running before the DHCP negotiation for the virtual adapter's IP starts.
|
|
NOTE: For distribution-specific issues, see ldistrib.htm. |
Always bind the protocol to the virtual adapter and NOT to the physical adapter for adapters that are in a team. Also, when changing settings such as the MTU or MAC address, always apply them to the virtual adapter. Intel ANS cannot prevent such operations, and failing to do as recommended might lead to unpredictable results.
If you receive the error message, "module not loaded," during configuration of the team, this means that the target module does not match the running kernel. Enter the iANS-x.x.x/src directory and run:
make veryclean
make install
If you have unsuccessfully attempted to add an adapter to a team with Jumbo Frames, check the adapter's ability to operate with Jumbo Frames.
If you receive the error message, "Intel Advanced Network Services (iANS) has determined you have selected network adapters that are not eligible to participate as members of an iANS team. Please install Intel Server Adapters in your system and re-install the advanced server features.", this indicates that either:
the team needs at least one Intel server adapter, or
at least one adapter is from another vendor and is not capable of participating in the team.
When trying to compile the driver by running make install, the following error may occur:
"Linux kernel source not configured - missing version.h"
To solve this issue, create the version.h file by going to the Linux source tree and doing a 'make include/linux/version.h'.
PCI Hot Plug Add/Remove.
When executing a PCI Hot Plug Remove operation on an adapter that is part of an ANS team, the adapter is automatically removed from the team.
When reactivating an adapter with a PCI Hot Plug Add operation, the new interface should be manually added back into the team by an ANS Hot Add operation.
If you are running an AFT/SFT/ALB team and receive the following messages numerous times, "Adapter ethx does not sense any probes. Possible reason: Adapter deactivated, Adapter isolated, Partitioned team." and then, "Adapter ethx rejoined.", probe packets are being dropped (by the kernel queues, switch, etc.) resulting in the ANS team losing probes. This could happen in a healthy system that is under heavy stress.
As a workaround, disable the probes by issuing the following command:
ianscfg -P -t <team_name> -d
Different switches react differently when the switch manager disables a switch port. Some switches react with a link loss of the disabled port. Others do not change the link status at all; they simply stop traffic. If the link status is not affected when a port is disabled then ANS cannot always detect that a switch port that is connected to an ANS member has been disabled by the switch manager. ANS detects that the port has been disabled even though no link status change occurs only if one of two conditions exist:
Probes are enabled and there are more than two enabled and active members in the team (relevant to AFT, SFT, and ALB teaming modes only). ANS detects that the member attached to the disabled port is missing probes while others are receiving and transmitting probes successfully.
Teaming mode is 802.3ad - ANS deactivates the member attached to the disabled port since no LACPDUs will be received by ANS on the disabled port.
To avoid problems of fault tolerance in FEC, GEC, and ALB teaming modes the switch manager should physically disconnect the link on the disabled port or remove the member connected to the disabled port from the team. In AFT and SFT modes it is enough to set an active member that is attached to an enabled port to be the preferred primary and make sure that the priority of the member attached to the disabled port is "none".
When setting a new MAC address for an AFT/ALB team (for the virtual adapter), if the new MAC address is already in use by a member of the team and that MAC address was not chosen as the team's MAC address upon committing the team, then traffic might become unstable. Intel ANS does not automatically check for duplicate MAC addresses. Therefore, before setting a new MAC address in this type of scenario, ensure that the intended MAC address is not the same as those used for members of the team.
If you load the ipv6 module when there are already
active ANS teams, then
ifconfig <ethX>
where
<ethX> is an ANS team member interface. For every IPv6 address that
is
inet6 addr: <addr>/<prefix len> Scope:<scope>
ifconfig <ethX> del <addr>/<prefix len>
If
the ipv6 module was already loaded in your system and you’ve manually
Multiple identical packets received by non-Intel adapters in AFT/ALB team: When connecting an AFT/ALB team with MVT members to a hub (and sometimes to a switch) the ethernet frames that are designated to the address of the team will be received by the primary member and all MVT members, without filtering. Possible lower throughput and higher cpu usage may occur. This happens because non-Intel adapters do not allow setting their MAC address after activation and therefore are set to the team's address by default.
Inconsistent Rx rebalancing when hot adding an adapter in RLB mode: When hot adding a member to an RLB team, Rx balancing may not occur immediately. This may be because ARP packets were lost or ignored by the clients. Re-balancing occurs periodically every 5 minutes; the new added member will share the load at that time.
When restoring the link for one of the adapters in a team, there may be some traffic loss if there is a forwarding delay period between the link up indication and the time at which the link is actually capable of forwarding traffic. The forwarding delay can vary from one system to another and in many cases is negligible. If your system does experience forward delay then you can configure an equivalent forward delay in ANS such that a member will not be activated until the delay is up. See the ianscfg man page for more information on how to configure forward delay.
Last modified on 2/29/04 12:29p Revision 8