NVIDIA NVOS User Manual for InfiniBand Switches

IB Router

InfiniBand (IB) Router provides the ability to send traffic between two or more IB subnets thereby potentially expanding the size of the network to over 40k end-ports, enabling separation, fault resilience and connection to different topologies used by different subnets.

For more information on IB router architecture and functionality, please refer to the community post IB Router Architecture and Functionality.

Configuring IB Router

Prerequisites

  1. Platform support: IB Router functionality is supported only on Q3400-RA and Q3401-RD systems.

  2. Subnet Manager (SM): Ensure Subnet Managers are not running before configuring the switch as an IB router and assigning interfaces to the IB subnets (SWIDs). Complete all IB router configuration and SWID assignments before starting SMs.

  3. Limitations:

    IPoIB: When IB router profile is configured, IPoIB interfaces are not supported and will remain down.

    FNM: External FNM ports are restricted to the default SWID and cannot be assigned to other subnets. Servers should not be connected to the external FNM ports on IB router switches.

NVOS Switch Configuration

  1. Configure the system profile to enable IB routing with multiple SWIDs:

    admin@nvos:~$ nv action change system profile ib-routing enabled num-of-swids 4
    The operation will reset the system configuration and initiate a reboot.
    The resulting system profile values will be:
    ib-routing: enabled
    num-of-swids: 4
    adaptive-routing-groups: 1792
    Type [y] to confirm the system profile change and reboot.
    Type [N] to abort.
    Do you want to continue? [y/N] y
    Action executing ...
    System will be rebooted, configurations and system files might be deleted
    Action executing ...
    

    Profile change triggers a factory reset. All existing configurations will be reset to defaults and the system will reboot.

  2. After the reboot, verify the system profile configuration:

    admin@nvos:~$ nv show system profile
                             operational
    -----------------------  -----------
    adaptive-routing         enabled
    adaptive-routing-groups  1792
    breakout-mode            enabled
    ib-routing               enabled
    num-of-swids             4
    
    • Ensure that ib-routing is set to enabled and that num-of-swids matches your configured value.

    • After the profile change, all IB interfaces are assigned to the default SWID (infiniband-default). To confirm the IB interface assignment to the infiniband-default SWID, use the nv show interface command.


  3. Assign each IB interface to the appropriate IB subnet (SWID) according to the topology port mappings.
    Use the following command format:

    admin@nvos:~$ nv set interface <interface-id> link ib-subnet <swid-name>
    

    where <interface-id> is an interface name or range, and <swid-name> is the target SWID.


    Example:

    admin@nvos:~$ nv set interface sw3-4p1-2 link ib-subnet infiniband-1
    admin@nvos:~$ nv config apply -y
    Warning: IB interface configuration is changed. In order to apply configuration the port will be restarted.
    

    Ensure Subnet Managers are not running before assigning interfaces to SWIDs. SWID assignments must be completed prior to SM configuration.
    Changing SWID assignments while Subnet Managers are running may cause inconsistent SM state requiring a reboot of the router.

  4. Verify the SWID (IB subnet) assignment for IB interfaces:

    admin@nvos:~$ nv show interface
    Interface  State  Speed  MTU    Type   IB Speed  IB Subnet           Logical State  Physical State
    ---------  -----  -----  -----  -----  --------  ------------------  -------------  --------------
    eth0       up     1G     1500   eth                                                                
    eth1       up     1G     1500   eth                                                                
    fnm1       up     800G   4096   fnm    xdr       infiniband-default  Active         LinkUp
    sw1p1      up     800G   4096   ib     xdr       infiniband-default  Active         LinkUp
    sw1p2      up     800G   4096   ib     xdr       infiniband-default  Active         LinkUp
    sw2p1      up     800G   4096   ib     xdr       infiniband-default  Active         LinkUp
    sw2p2      up     800G   4096   ib     xdr       infiniband-default  Active         LinkUp
    sw3p1      up     800G   4096   ib     xdr       infiniband-1        Active         LinkUp
    sw3p2      up     800G   4096   ib     xdr       infiniband-1        Active         LinkUp
    sw4p1      up     800G   4096   ib     xdr       infiniband-1        Active         LinkUp
    sw4p2      up     800G   4096   ib     xdr       infiniband-1        Active         LinkUp
    ...
    

Verify IB router

After configuring both the IB router and starting Subnet Managers, verify the routing table:

admin@nvos:~$ nv show ib router routing-table

IB Subnet           Subnet Prefix            Valid

------------------  -----------------------  -------

infiniband-default  fe:c0:00:00:00:00:00:01  true

infiniband-1        fe:c0:00:00:00:00:00:02  true
infiniband-2        00:00:00:00:00:00:00:00  false
infiniband-3        00:00:00:00:00:00:00:00  false

To view the router counters:

admin@nvos:~$ nv show ib router counters
Counter                       infiniband-default  infiniband-1  infiniband-2  infiniband-3
----------------------------  ------------------  ------------  ------------  ------------
in-bytes                      0                   0             0             0
in-pkts                       0                   0             0             0
in-drops                      0                   0             0             0
out-bytes                     0                   0             0             0
out-pkts                      0                   0             0             0
out-drops                     0                   0             0             0
out-errors                    0                   0             0             0
out-unicast-pkts              0                   0             0             0
out-multicast-pkts            0                   0             0             0
in-unicast-pkts               0                   0             0             0
in-multicast-pkts             0                   0             0             0
buffer-overrun-errors         0                   0             0             0
out-wait                      0                   0             0             0
carrier-down-count            0                   0             0             0
error-recovery                0                   0             0             0
port-rcv-switch-relay-errors  0                   0             0             0
port-rcv-constraint-errors    0                   0             0             0
local-integrity-errors        0                   0             0             0



IB Router Commands



Last updated: