Transceiver and Cable Self-qualification with Cumulus Linux

NVIDIA provides this document as a suggested procedure for qualifying a transceiver module or cable as a recommended pluggable. NVIDIA does not actively prevent any non-recommended pluggable from functioning and does not restrict the use of self-qualified pluggables. Customers who wish to use a non-recommended pluggable can follow the suggested procedure outlined in this document. For concerns about pluggables that are not listed, contact your NVIDIA sales team.

Example Connections

This procedure is valid for qualifying all types of transceivers and cables in a device undergoing testing. Customers can choose the same types or a combination of different transceivers for this test.

The following diagram illustrates an example where the top and bottom ports connect with a cable as a loopback. The example includes cabling and configuration for the testing of both 40G QSFP and 10/1G SFP. NVIDIA recommends that you test different speed components independently. The example includes both to simplify the presentation.

Loopback cable port connections

TG-1/TG-2 are either networking traffic generators (IXIA/Spirent or equivalent) or two servers with a Linux OS installed and iperf3 free traffic generator tool or an equivalent.

Cumulus Linux Configuration

Modify the /etc/network/interfaces file with the following bridge configuration:

auto trg_1
iface trg_1 inet manual
    bridge_ageing 150
    bridge_stp off
    bridge_ports swp41 swp52
    up ip link set trg_1 up

auto trg_2
iface trg_2 inet manual
    bridge_ageing 150
    bridge_stp off
    bridge_ports swp42 swp43
    up ip link set trg_2 up

auto l_1
iface l_1 inet manual
    bridge_ageing 150
    bridge_stp off
    bridge_ports swp51 swp50
    up ip link set l_1 up

auto l_2
iface l_2 inet manual
    bridge_ageing 150
    bridge_stp off
    bridge_ports swp49 swp48
    up ip link set l_2 up

auto l_3
iface l_3 inet manual
    bridge_ageing 150
    bridge_stp off
    bridge_ports swp47 swp46
    up ip link set l_3 up

auto l_4
iface l_4 inet manual
    bridge_ageing 150
    bridge_stp off
    bridge_ports swp45 swp44
    up ip link set l_4 up

You should expect to see the following bridge configuration after you reboot the switch you are testing. Rebooting instead of reloading the configuration ensures that the system detects and properly configures all optics when the it starts up.

cumulus@switch~$ sudo brctl show
bridge name    bridge id                 STP enabled       interfaces
l_1            8000.443839002076         no                swp50
l_2            8000.443839002071         no                swp48
l_3            8000.44383900206f         no                swp46
L_4            8000.44383900206e         no                swp44
trg_1          8000.44383900206c         no                swp43
trg_2          8000.44383900206d         no                swp44

Run the following Cumulus Linux command to verify that all loopback links are up:

~$ sudo ethtool swp50
Settings for swp50:
               Supported ports: [ FIBRE ]
               Supported link modes:   10000baseT/Full
               Supported pause frame use: Symmetric Receive-only
               Supports auto-negotiation: Yes
               Advertised link modes:  1000baseT/Full
               Advertised pause frame use: Symmetric
               Advertised auto-negotiation: No
               Speed: 40000Mb/s
               Duplex: Full
               Port: FIBRE
               PHYAD: 0
               Transceiver: external
               Auto-negotiation: off
               Current message level: 0x00000000 (0)
               Link detected: yes

Activating Data Traffic from Linux Servers

Configure the server interfaces connected to traffic ingress/egress ports to set IPv4 addresses in the same IP subnet. For example:

TG1$ sudo ifconfig eth1 up
TG2$ sudo ifconfig eth1 up

You can try ping or ping –f (flood) between these interfaces.

For generating iperf traffic, use options like the following (for example):

TG1$ sudo iperf -s -B -p 9000
TG2$ sudo iperf -c -i 3 -t 600 -p 9000 –d


  • -B is bound to an interface
  • -p is the TCP port number
  • -c is the iperf destination
  • -i is the print to screen interval
  • -t is the duration of the test in seconds
  • -d is bidirectional traffic

Confirm that ping and iperf traffic reach the destination and the bandwidth matches the expected rate (subject to the transceiver’s supported speed and the server CPU). Connect two servers back-to-back first to capture baseline server performance characteristics.

Transceivers Module Information (EEPROM & DOM)

Use the following Cumulus Linux command to check each transceiver’s EEPROM and Digital Optical Monitoring (DOM) information:

cumulus@switch~$ sudo ethtool –m swp<id>

Cumulus Linux-based Error Counters Check

The following commands indicate error and drop counters occurred during and after the test:

cumulus@switch~$ sudo ethtool -S swp<id> | grep -i error
     HwIfInDot3LengthErrors: 0
     HwIfInErrors: 0
     SoftInErrors: 0
     SoftInFrameErrors: 0
     HwIfOutErrors: 0
     SoftOutErrors: 0

PASS/FAIL Criteria

The following checklist and test plan comprise successful results.


Expected Result


$ ping –f <dest IP>
% packets drop

Source and destination IP of TG1 & TG2

$ iperf TCP stream

Bidirectional traffic with Cumulus Linux snake test matches transfer rate of two traffic generators' endpoints when connected back-to-back.

$ ethtool -S swp<id> | grep -i error

All error counters return null.

To detects drops or errors, collect output from ethtool -S before sending traffic, then again afterwards and compare to see if the counters increase.

$ ethtool swp<id>

Link up

$ ethtool -m swp<id>

Returns EEPROM information, such as vendor and equipment type.

DOM information is optional for successful self-qualification of transceivers. In some cases ODM vendors elect to restrict information programmed for DOM.

$ sudo systemctl reset-failed switchd.service
$ sudo systemctl restart switchd.service

After the switchd service restarts, all links are back up and traffic recovers.

This test confirms that a switchd daemon restart does correctly program all required transceiver settings.

Reboot the switch and repeat same tests and checkpoints again.

All checks/tests iterations are successful.

During the qualification cycle for some transceivers, NVIDIA observed that marginal and disqualified transceivers exhibited failures with 10-25% failure rates across switch reboots.