Switch Port Attributes

Cumulus Linux exposes network interfaces for several types of physical and logical devices:

• lo is the network loopback device
• eth is a switch management port (for out of band management only)
• swp is a switch front panel port
• (optional) br is a bridge (IEEE 802.1Q VLAN)
• (optional) bond is a bond (IEEE 802.3ad link aggregation trunk, or port channel)

Each physical network interface (port) has several settings:

• Auto-negotiation
• Duplex Mode
• Link speed
• MTU
• FEC

For NVIDIA Spectrum ASICs, the firmware configures FEC, link speed, duplex mode and auto-negotiation automatically, following a predefined list of parameter settings until the link comes up. You can disable FEC if necessary, which forces the firmware to not try any FEC options.

MTU

Interface MTU applies to traffic traversing the management port, front panel or switch ports, bridge, VLAN subinterfaces, and bonds (both physical and logical interfaces). MTU is the only interface setting that you must set manually.

In Cumulus Linux, ifupdown2 assigns 9216 as the default MTU setting. The initial MTU value set by the driver is 9238. After you configure the interface, the default MTU setting is 9216.

To change the MTU setting, run the following commands. The example command sets the MTU to 1500 for the swp1 interface.

cumulus@switch:~$ nv set interface swp1 link mtu 1500
cumulus@switch:~$ nv config apply

cumulus@switch:~$ sudo nano /etc/network/interfaces

auto swp1
iface swp1
    mtu 1500

cumulus@switch:~$ sudo ifreload -a

cumulus@switch:~$ sudo ip link set dev swp1 mtu 1500

Set a Global Policy

To set a global MTU policy, create a policy document (called mtu.json). For example:

cumulus@switch:~$ sudo cat /etc/network/ifupdown2/policy.d/mtu.json
{
  "address": {"defaults": { "mtu": "9216" }
            }
}

Bridge MTU

The MTU setting is the lowest MTU of any interface that is a member of the bridge (every interface specified in bridge-ports in the bridge configuration of the /etc/network/interfaces file). You are not required to specify an MTU on the bridge. Consider this bridge configuration:

auto bridge
iface bridge
    bridge-ports bond1 bond2 bond3 bond4 peer5
    bridge-vids 100-110
    bridge-vlan-aware yes

For a bridge to have an MTU of 9000, set the MTU for each of the member interfaces (bond1 to bond 4, and peer5) to 9000 at minimum.

When configuring MTU for a bond, configure the MTU value directly under the bond interface; the member links or slave interfaces inherit the configured value. If you need a different MTU on the bond, set it on the bond interface, as this ensures the slave interfaces pick it up. You do not have to specify an MTU on the slave interfaces.

VLAN interfaces inherit their MTU settings from their physical devices or their lower interface; for example, swp1.100 inherits its MTU setting from swp1. Therefore, specifying an MTU on swp1 ensures that swp1.100 inherits the MTU setting for swp1.

If you are working with VXLANs, the MTU for a virtual network interface (VNI must be 50 bytes smaller than the MTU of the physical interfaces on the switch, as various headers and other data require those 50 bytes. Also, consider setting the MTU much higher than 1500.

To show the MTU setting for an interface:

cumulus@switch:~$ nv show interface swp1

cumulus@switch:~$ ip link show dev swp1
3: swp1: <BROADCAST,MULTICAST,UP,LOWER_UP> mtu 9216 qdisc pfifo_fast state UP mode DEFAULT qlen 500
   link/ether 44:38:39:00:03:c1 brd ff:ff:ff:ff:ff:ff

Drop Packets that Exceed the Egress Layer 3 MTU

The switch forwards all packets that are within the MTU value set for the egress layer 3 interface. However, when packets are larger in size than the MTU value, the switch fragments the packets that do not have the DF bit set and drops the packets that do have the DF bit set.

Run the following command to drop all IP packets that are larger in size than the MTU value for the egress layer 3 interface instead of fragmenting packets:

cumulus@switch:~$ nv set system control-plane trap l3-mtu-err state off
cumulus@switch:~$ nv config apply

cumulus@switch:~$ echo "0 >" /cumulus/switchd/config/trap/l3-mtu-err/enable

FEC

FEC is an encoding and decoding layer that enables the switch to detect and correct bit errors introduced over the cable between two interfaces. The target IEEE BER on high speed Ethernet links is 10^-12. Because 25G transmission speeds can introduce a higher than acceptable BER on a link, FEC is often required to correct errors to achieve the target BER at 25G, 4x25G, 100G, and higher link speeds. The type and grade of a cable or module and the medium of transmission determine which FEC setting is necessary.

For the link to come up, the two interfaces on each end must use the same FEC setting.

The two FEC types are:

• Reed Solomon (RS), IEEE 802.3 Clause 108 (CL108) on individual 25G channels and Clause 91 on 100G (4channels). This is the highest FEC algorithm, providing the best bit-error correction.
• Base-R (BaseR), Fire Code (FC), IEEE 802.3 Clause 74 (CL74). Base-R provides less protection from bit errors than RS FEC but adds less latency.

Cumulus Linux includes additional FEC options:

• Auto FEC instructs the hardware to select the best FEC. For copper DAC, the remote end can negotiate FEC. However, optical modules do not have auto-negotiation capability; if the device chooses a preferred mode, it might not match the remote end. This is the current default on the NVIDIA Spectrum switch.
• No FEC (no error correction).

For 25G DAC, 4x25G Breakouts DAC and 100G DAC cables, the IEEE 802.3by specification creates 3 classes:

• CA-25G-L (Long cable) - Requires RS FEC - Achievable cable length of at least 5m. dB loss less or equal to 22.48. Expected BER of 10^-5 or better without RS FEC enabled.
• CA-25G-S (Short cable) - Requires Base-R FEC - Achievable cable length of at least 3m. dB loss less or equal to 16.48. Expected BER of 10^-8 or better without Base-R FEC enabled.
• CA-25G-N (No FEC) - Does not require FEC - Achievable cable length of at least 3m. dB loss less or equal to 12.98. Expected BER 10^-12 or better with no FEC enabled.

The IEEE classification specifies various dB loss measurements and minimum achievable cable length. You can build longer and shorter cables if they comply to the dB loss and BER requirements.

If a cable has a CA-25G-S classification and FEC is not on, the BER might be unacceptable in a production network. It is important to set the FEC according to the cable class (or better) to have acceptable bit error rates. See Determining Cable Class below.

You can check bit errors using cl-netstat (RX_ERR column) or ethtool -S (HwIfInErrors counter) after a large amount of traffic passes through the link. A non-zero value indicates bit errors. Expect error packets to be zero or extremely low compared to good packets. If a cable has an unacceptable rate of errors with FEC enabled, replace the cable.

For 25G, 4x25G Breakout, and 100G Fiber modules and AOCs, there is no classification of 25G cable types for dB loss, BER or length. Use FEC if the BER is low enough.

Cable Class of 100G and 25G DACs

You can determine the cable class for 100G and 25G DACs from the Extended Specification Compliance Code field (SFP28: 0Ah, byte 35, QSFP28: Page 0, byte 192) in the cable EEPROM programming.

For 100G DACs, most manufacturers use the 0x0Bh 100GBASE-CR4 or 25GBASE-CR CA-L value (the 100G DAC specification predates the IEEE 802.3by 25G DAC specification). Use RS FEC for 100G DAC; shorter or better cables might not need this setting.

You can determine for which grade the manufacturer has designated the cable as follows.

For the SFP28 DAC, run the following command:

cumulus@switch:~$ sudo ethtool -m swp1 hex on | grep 0020 | awk '{ print $6}'
0c

The values at location 0x0024 are:

• 0x0b : CA-L (long cable - RS FEC required)
• 0x0c : CA-S (short cable - Base-R or better FEC required)
• 0x0d : CA-N (no FEC required)

For the QSFP28 DAC, run the following command:

cumulus@switch:~$ sudo ethtool -m swp1s0 hex on | grep 00c0 | awk '{print $2}'
0b

The values at 0x00c0 are:

• 0x0b : CA-L (long cable - RS FEC required) or 100G CR4
• 0x0c : CA-S (short cable - Base-R or better FEC required)
• 0x0d : CA-N (no FEC required)

In each example below, the Compliance field comes from the method described above; the ethool -m output does not show it.

3meter cable that does not require FEC
(CA-N)
Cost: More expensive
Cable size: 26AWG (Note that AWG does not necessarily correspond to overall dB loss or BER performance)
Compliance Code: 25GBASE-CR CA-N

3meter cable that requires Base-R FEC
(CA-S)
Cost: Less expensive
Cable size: 26AWG
Compliance Code: 25GBASE-CR CA-S

When in doubt, consult the manufacturer directly to determine the cable classification.

Spectrum ASIC FEC Behavior

The firmware in a Spectrum ASIC applies FEC configuration to 25G and 100G cables based on the cable type and whether the peer switch also has a Spectrum ASIC.

When the link is between two switches with Spectrum ASICs:

• For 25G optical modules, the Spectrum ASIC firmware chooses Base-R/FC-FEC.
• For 25G DAC cables with attenuation less or equal to 16db, the firmware chooses Base-R/FC-FEC.
• For 25G DAC cables with attenuation higher than 16db, the firmware chooses RS-FEC.
• For 100G cables/modules, the firmware chooses RS-FEC.

When linking to a non-Spectrum peer, the firmware lets the peer decide. The Spectrum ASIC supports RS-FEC (for both 100G and 25G), Base-R/FC-FEC (25G only), or no-FEC (for both 100G and 25G).

How Does Cumulus Linux use FEC?

A Spectrum switch enables FEC automatically when it powers up. The port firmware tests and determines the correct FEC mode to bring the link up with the neighbor. It is possible to get a link up to a switch without enabling FEC on the remote device as the switch eventually finds a working combination to the neighbor without FEC.

The following sections describe how to show the current FEC mode, and how to enable and disable FEC.

Show the Current FEC Mode

To show the FEC mode on a switch port, run the NVUE nv show interface <interface> link command.

cumulus@switch:~$ nv show interface swp1 link
                  operational   applied  pending  description
----------------  ------------  -------  -------  ----------------------------------------------------------------------
auto-negotiate    off           on       on       Link speed and characteristic auto negotiation
breakout                        1x       1x       sub-divide or disable ports (only valid on plug interfaces)
duplex            full          full     full     Link duplex
fec                             auto     auto     Link forward error correction mechanism
...

Enable or Disable FEC

To enable Reed Solomon (RS) FEC on a link:

cumulus@switch:~$ nv set interface swp1 link fec rs
cumulus@switch:~$ nv config apply

cumulus@switch:~$ sudo nano /etc/network/interfaces

auto swp1
iface swp1
    link-autoneg off
    link-speed 100000
    link-fec rs

cumulus@switch:~$ sudo ifreload -a

cumulus@switch:~$ sudo ethtool --set-fec swp1 encoding RS

To enable Base-R/FireCode FEC on a link:

cumulus@switch:~$ nv set interface swp1 link fec baser
cumulus@switch:~$ nv config apply

cumulus@switch:~$ sudo nano /etc/network/interfaces

auto swp1
iface swp1
    link-autoneg off
    link-speed 100000
    link-fec baser

cumulus@switch:~$ sudo ifreload -a

cumulus@switch:~$ sudo ethtool --set-fec swp1 encoding BaseR

To enable FEC with Auto-negotiation:

cumulus@switch:~$ nv set interface swp1 link auto-negotiate on
cumulus@switch:~$ nv config apply

cumulus@switch:~$ sudo nano /etc/network/interfaces

auto swp1
iface swp1
link-autoneg on

cumulus@switch:~$ sudo ifreload -a

ethtool -s swp1 speed 10000 duplex full autoneg on

To show the FEC and auto-negotiation settings for an interface, either run the NVUE nv show interface <interface> link command or the Linux sudo ethtool swp1 | egrep 'FEC|auto' command:

cumulus@switch:~$ sudo ethtool swp1 | egrep 'FEC|auto'
Supports auto-negotiation: Yes
Supported FEC modes: RS
Advertised auto-negotiation: Yes
Advertised FEC modes: RS
Link partner advertised auto-negotiation: Yes
Link partner advertised FEC modes: Not reported

To disable FEC on a link:

cumulus@switch:~$ nv set interface swp1 link fec off
cumulus@switch:~$ nv config apply

cumulus@switch:~$ sudo nano /etc/network/interfaces

auto swp1
iface swp1
link-fec off

cumulus@switch:~$ sudo ifreload -a

cumulus@switch:~$ sudo ethtool --set-fec swp1 encoding off

DR1 and DR4 Modules

100GBASE-DR1 modules, such as NVIDIA MMS1V70-CM, include internal RS FEC processing, which the software does not control. When using these optics, you must either set the FEC setting to off or leave it unset for the link to function.

400GBASE-DR4 modules, such as NVIDIA MMS1V00-WM, require RS FEC. The switch automatically enables FEC if it is set to off.

You typically use these optics to interconnect 4x SN2700 uplinks to a single SN4700 breakout downlink. The following configuration shows an explicit FEC example. You can leave the FEC settings unset for autodetection.

SN4700 (400GBASE-DR4 in swp1):

cumulus@SN4700:mgmt:~$ nv set interface swp1 link breakout 4x lanes-per-port 2
cumulus@SN4700:mgmt:~$ nv set interface swp1s0 link fec rs
cumulus@SN4700:mgmt:~$ nv set interface swp1s0 link speed 100G
cumulus@SN4700:mgmt:~$ nv set interface swp1s1 link fec rs
cumulus@SN4700:mgmt:~$ nv set interface swp1s1 link speed 100G
cumulus@SN4700:mgmt:~$ nv set interface swp1s2 link fec rs
cumulus@SN4700:mgmt:~$ nv set interface swp1s2 link speed 100G
cumulus@SN4700:mgmt:~$ nv set interface swp1s3 link fec rs
cumulus@SN4700:mgmt:~$ nv set interface swp1s3 link speed 100G
cumulus@SN4700:mgmt:~$ nv config apply

SN2700 (100GBASE-DR1 in swp11-14):

cumulus@SN2700:mgmt:~$ nv set interface swp11 link fec off
cumulus@SN2700:mgmt:~$ nv set interface swp11 link speed 100G
cumulus@SN2700:mgmt:~$ nv set interface swp12 link fec off
cumulus@SN2700:mgmt:~$ nv set interface swp12 link speed 100G
cumulus@SN2700:mgmt:~$ nv set interface swp13 link fec off
cumulus@SN2700:mgmt:~$ nv set interface swp13 link speed 100G
cumulus@SN2700:mgmt:~$ nv set interface swp14 link fec off
cumulus@SN2700:mgmt:~$ nv set interface swp14 link speed 100G
cumulus@SN4700:mgmt:~$ nv config apply

The FEC operational view of this configuration appears incorrect because FEC is operationally enabled only on the SN4700 400G breakout side. This is because the 100G DR1 module side handles FEC internally, which is not visible to Cumulus Linux.

cumulus@SN2700:mgmt:~$ nv show int swp11 link
                       operational        applied
---------------------  -----------------  -------
auto-negotiate         on                 on     
duplex                 full               full   
speed                  100G               auto   
fec                    off                off   
mtu                    9216               9216   
fast-linkup            off                       
[breakout]                                       
state                  up                 up     
...

cumulus@SN4700:mgmt:~$ nv show int swp1s1 link
                       operational        applied
---------------------  -----------------  -------
auto-negotiate         on                 on     
duplex                 full               full   
speed                  100G               auto   
fec                    rs                 off    
mtu                    9216               9216   
fast-linkup            off                       
[breakout]                                       
state                  up                 up     
...

Default Policies for Interface Settings

Instead of configuring settings for each individual interface, you can specify a policy for all interfaces on a switch or tailor custom settings for each interface. Create a file in /etc/network/ifupdown2/policy.d/ and populate the settings accordingly. The following example shows a file called address.json.

cumulus@switch:~$ cat /etc/network/ifupdown2/policy.d/address.json
{
    "ethtool": {
        "defaults": {
            "link-duplex": "full"
        },
        "iface_defaults": {
            "swp1": {
                "link-autoneg": "on",
                "link-speed": "1000"
          },
            "swp16": {
                "link-autoneg": "off",
                "link-speed": "10000"
            },
            "swp50": {
                "link-autoneg": "off",
                "link-speed": "100000",
                "link-fec": "rs"
            }
        }
    },
    "address": {
        "defaults": { "mtu": "9000" },
        "iface_defaults": {
            "eth0": {"mtu": "1500"}
        }
    }
}

Breakout Ports

Cumulus Linux supports the following ports breakout options:

Configure a Breakout Port

You can break out (split) a port using the following options:

• 1x does not split the port. This is the default port setting.
• 2x splits the port into two interfaces.
• 4x splits the port into four interfaces.
• 8x splits the port into eight interfaces.

If you split a 100G port into four interfaces and auto-negotiation is on (the default setting), Cumulus Linux advertises the speed for each interface up to the maximum speed possible for a 100G port (100/4=25G). You can overide this configuration and set specific speeds for the split ports if necessary.

cumulus@switch:~$ nv set interface swp1 link breakout 4x
cumulus@switch:~$ nv set interface swp1s0-3 link state up
cumulus@switch:~$ nv config apply

cumulus@switch:~$ nv set interface swp1 link breakout 4x
cumulus@switch:~$ nv set interface swp1s0-3 link state up
cumulus@switch:~$ nv set interface swp1s0-3 link speed 10G

cumulus@switch:~$ sudo cat /etc/network/interfaces
...
auto swp1s0
iface swp1s0
    link-speed 10000 
    link-duplex full 
    link-autoneg off
auto swp1s1
iface swp1s1
    link-speed 10000 
    link-duplex full 
    link-autoneg off
auto swp1s2
iface swp1s2
    link-speed 10000 
    link-duplex full 
    link-autoneg off
auto swp1s3
iface swp1s3
    link-speed 10000 
    link-duplex full 
    link-autoneg off
...

cumulus@switch:~$ sudo ifreload -a

The SN4700 and SN4410 switch does not support auto-negotiation on QSFP-DD 400G transceiver modules. You need to force set the speed.

Set the Number of Lanes per Split Port

By default, to calculate the split port width, Cumulus Linux uses the formula split port width = full port width / breakout. For example, a port split into two interfaces (2x breakout) => 8 lanes width / 2x breakout = 4 lanes per split port.

If you need to use a different port width than the default, you can set the number of lanes per port.

The following example command splits swp1 into two interfaces (2x) and sets the number of lanes per split port to 2.

cumulus@switch:~$ nv set interface swp1 link breakout 2x lanes-per-port 2
cumulus@switch:~$ nv config apply

cumulus@switch:~$ nv unset interface swp1 link breakout
cumulus@switch:~$ nv config apply
cumulus@switch:~$ nv set interface swp1 link breakout 2x lanes-per-port 2
cumulus@switch:~$ nv config apply

cumulus@switch:~$ sudo nano /etc/cumulus/ports_width.conf
...
1=2
2=default
3=default
4=default
5=default
6=default
7=default
8=default
...

cumulus@switch:~$ sudo systemctl reload switchd.service

Remove a Breakout Port

To remove a breakout port:

Configure Port Lanes

You can override the default behavior for supported speeds and platforms and specify the number of lanes for a port. For example, for the NVIDIA SN4700 switch, the default port speed is 50G (2 lanes, NRZ signaling mode) and 100G (4 lanes, NRZ signaling mode). You can override this setting to 50G (1 lane, PAM4 signaling mode) and 100G (2 lanes, PAM4 signaling mode).

cumulus@switch:~$ nv set interface swp1 link speed 50G
cumulus@switch:~$ nv set interface swp1 link lanes 1
cumulus@switch:~$ nv config apply 

cumulus@switch:~$ nv set interface swp2 link speed 100G
cumulus@switch:~$ nv set interface swp2 link lanes 2
cumulus@switch:~$ nv config apply

ports.conf File Validator

Cumulus Linux includes a ports.conf validator that switchd runs automatically before the switch starts up to confirm that the file syntax is correct. You can run the validator manually to verify the syntax of the file whenever you make changes. The validator is useful if you want to copy a new ports.conf file to the switch with automation tools, then validate that it has the correct syntax.

To run the validator manually, run the /usr/cumulus/bin/validate-ports -f <file> command. For example:

cumulus@switch:~$ /usr/cumulus/bin/validate-ports -f /etc/cumulus/ports.conf

Troubleshooting

This section shows basic commands for troubleshooting switch ports. For a more comprehensive troubleshooting guide, see Troubleshoot Layer 1.

Statistics

To show interface statistics, run the NVUE nv show interface <interface> stats command or the Linux sudo ethtool -S <interface> command.

cumulus@switch:~$ nv show interface swp1 stats
                    operational  applied
-------------------  -----------  -------
carrier-transitions  4                   
in-bytes             3.37 MB             
in-drops             0                   
in-errors            0                   
in-pkts              29025               
out-bytes            4.28 MB             
out-drops            0                   
out-errors           0                   
out-pkts             43945  
...

Query SFP Port Information

To verify SFP settings, run the ethtool -m command. The following example shows the vendor, type and power output for swp1.

cumulus@switch:~$ sudo ethtool -m swp1 | egrep 'Vendor|type|power\s+:'
        Transceiver type                          : 10G Ethernet: 10G Base-LR
        Vendor name                               : FINISAR CORP.
        Vendor OUI                                : 00:90:65
        Vendor PN                                 : FTLX2071D327
        Vendor rev                                : A
        Vendor SN                                 : UY30DTX
        Laser output power                        : 0.5230 mW / -2.81 dBm
        Receiver signal average optical power     : 0.7285 mW / -1.38 dBm

Considerations

Important Upgrade Information for Breakout Ports and NVUE

Cumulus Linux 5.4 and later uses a new format for port splitting but continues to support the old port split configuration in the /etc/cumulus/ports.conf file. However, NVUE has deprecated the port split command options (2x10G, 2x25G, 2x40G, 2x50G, 2x100G, 2x200G, 4x10G, 4x25G, 4x50G, 4x100G, 8x50G) available in Cumulus Linux 5.3 and earlier, with no backwards compatibility. If you used NVUE to configure port breakout speeds in Cumulus 5.3 or earlier, you must either:

• Upgrade to Cumulus Linux 5.4 with a binary installation, then configure port breakouts on your switch with the new NVUE syntax.
• Upgrade to Cumulus Linux 5.4 with package installation (apt upgrade) but follow the steps below before you upgrade.

To upgrade with apt upgrade:

1. Before you run apt upgrade, save the current NVUE configuration with the nv config save command.
2. Edit the /etc/nvue.d/startup.yaml file to use the new NVUE breakout syntax for each breakout port. For example, change 4x10G to 4x: {}:

Change:

interface:
  swp1:
    link:
      breakout: 4x10G
    type: swp
...
   swp2:
     link:
       breakout: disabled
     type: swp

to

interface:
  swp1:
    link:
      breakout:
        4x: {}
    type: swp
...
   swp2:
     link:
       breakout:
         disabled: {}
     type: swp

3. Upgrade to Cumulus Linux 5.4 with apt-get upgrade.

Auto-negotiation and FEC

If auto-negotiation is off on 100G and 25G interfaces, you must set FEC to OFF, RS, or BaseR to match the neighbor. The FEC default setting of auto does not link up when auto-negotiation is off.

Auto-negotiation and Link Speed

If auto-negotiation is on and you set the link speed for a port, Cumulus Linux disables auto-negotiation and uses the port speed setting you configure.

1000BASE-T SFP Modules Supported Only on Certain 25G Platforms

The following 25G switches support 1000BASE-T SFP modules:

• NVIDIA SN2410
• NVIDIA SN2010

100G or faster switches do not support 1000BASE-T SFP modules.

NVIDIA SN2100 Switch and eth0 Link Speed

After rebooting the NVIDIA SN2100 switch, eth0 always has a speed of 100MB per second. If you bring the interface down and then back up again, the interface negotiates 1000MB. This only occurs the first time the interface comes up.

To work around this issue, add the following commands to the /etc/rc.local file to flap the interface automatically when the switch boots:

modprobe -r igb
sleep 20
modprobe igb

Delay in Reporting Interface as Operational Down

When you remove two transceivers simultaneously from a switch, both interfaces show the carrier down status immediately. However, it takes one second for the second interface to show the operational down status. In addition, the services on this interface also take an extra second to come down.

NVIDIA Spectrum-2 Switches and FEC Mode

The NVIDIA Spectrum-2 (25G) switch only supports RS FEC.

Related Information

• Debian - Network Configuration