NVIDIA Scalable Hierarchical Aggregation and Reduction Protocol (SHARP) Rev 3.8.0
NVIDIA Scalable Hierarchical Aggregation and Reduction Protocol (SHARP) Rev 3.8.0

Operating NVIDIA SHARP with PKeys

SHARP can operate in a system that has either a single static special PKey, or a system that dynamically allocates PKeys.

This method is used when SHARP is intended to operate exclusively on a single known PKey. To implement this, adjust the ib_qpc_pkey field to the desired PKey value in the SHARP configuration file.

Remember to ensure the membership bit is properly set, which entails setting both bit 0x8000 and the corresponding pkey value at all times.

SHARP enables dynamic declaration of PKeys when operating from the UFM management host.

This feature is facilitated by the enable_sharp_allocation config parameter mentioned in the SHARP Reservation section. Configuring SHARP to operate in reservation mode via the UFM config file allows UFM to relay PKeys information to SHARP.

  • To enable this functionality, make sure to set the following parameter in the UFM gv.cfg file:

    sharp_allocation_enabled = true

  • Restart UFM to apply the updated settings.

Note that the manipulation of PKeys does not automatically update the SHARP configuration, meaning that there are 2 sets of REST-API calls that need to be used:

  1. The pure PKey REST API : This API affects the PKEY settings, however, it is not related to SHARP.

    Note

    See the PKey GUIDs REST API section in the latest NVIDIA UFM Enterprise REST API guide here.

  2. The SHARP Reservation REST API: This API sets SHARP reservation and can refer to the pkeys that were created in the previous API.

    Note

    See the NVIDIA SHARP REST API section in the latest NVIDIA UFM Enterprise REST API guide here.

© Copyright 2024, NVIDIA. Last updated on Aug 13, 2024.