Reducing the Number of Operational VLs in the Subnet in OpenSM

When streaming aggregation is enabled, it is recommended to reduce the number of enabled VLs in the subnet to 2 VLs.

To reduce the number of VLs, max_op_vl parameter of OpenSM has to be set according to required number of VLs. For example, configure OpenSM to enable 2 VLs in the subnet by changing the max_op_vl parameter in the OpenSM configuration file:

            

            
# Limit the maximal operational VLs
max_op_vls 2
        

By default, for maximal performance, NVIDIA Quantum switches support single SHARP streaming aggregation flow.

To enable two simultaneous SHARP streaming aggregation flows, please contact Support.

Disabling SHARP on a specific network switch device can be performed using OpenSM device configuration file by performing the following:

1. Define a port group with the specified network device in the port groups file. The group is specified by the pgrp_policy_file parameter in the OpenSM configuration file.
   For example:

   Copy

   Copied!

               
   
               
   port-group
   name: NON_SHARP_SWITCHES
   port-guid: 0x0002c90000000001
   end-port-group
           

2. Configure OpenSM to disable SHARP on the devices of the specified port groups in device configuration file specified by the device_configuration_file parameter in OpenSM configuration file.

   Copy

   Copied!

               
   
               
   	port-conf
   		port-group-name: NON_SHARP_SWITCHES
   		sharp-enabled: 0
   	end-port-conf
           

3. Reload OpenSM.