Server Vendor Partner Validation Guide

Subject: Graid Driver & System Compatibility Quick Validation Procedure
Target Audience: System Integrators and Server Vendor Partners

Introduction

This document outlines the standard procedure to validate the compatibility and stability of the Graid Technology SupremeRAID™ AE driver on new server systems. The following tests verify GPU recognition, service status, RAID operations, and data integrity under failure and recovery scenarios.

Prerequisites

• Root Privileges: All commands must be executed with sudo or as the root user.
• Tools Required: nvidia-smi, fio, graidctl.
• Target Drive: Ensure a Graid virtual drive is created and available (referenced herein as /dev/gdg0n1).

Environment & Driver Verification

Step 1: Verify GPU Detection
Use the NVIDIA System Management Interface to list all available GPUs and ensure the hardware is physically detected.

sudo nvidia-smi

Verification:

Ensure the Graid-compatible GPU appears in the device list.

Step 2: Check Fabric Manager Service
Verify that the NVIDIA Fabric Manager service is active and running.

sudo systemctl status nvidia-fabricmanager

Step 3: Verify Fabric Manager Registration
Confirm that the specific GPU is properly registered with the Fabric Manager.

sudo nvidia-smi -i 0 -q | grep -A 3 "^    Fabric"

Verification:

The output should indicate a successful registration status.

RAID Creation & Baseline Integrity Test

Step 4: Create Virtual Drive
Initialize a new Graid virtual drive using your standard configuration parameters.

Step 5: Baseline Data Generation (Write Test)
Assuming the virtual drive is located at /dev/gdg0n1, execute the following FIO command. This writes data to the drive and saves the checksum state for later verification.

sudo fio --filename=/dev/gdg0n1 \
  --ioengine=libaio --direct=1 --name=verify \
  --rw=write --bs=1048576 --numjobs=10 --iodepth=32 \
  --size=10% --offset_increment=10% \
  --cpus_allowed_policy=split \
  --verify=crc32c --verify_fatal=1 --verify_dump=1 \
  --verify_state_save=1 --group_reporting=1

Fault Injection & Degraded State Validation

Step 6: Simulate Drive Failure
Manually force a physical drive into a "Bad" state to degrade the RAID array.

sudo graidctl edit physical_drive 0 marker bad

Verification: Confirm the Drive Group status changes from OPTIMAL to DEGRADED or PARTIALLY_DEGRADED.

Step 7: Degraded State Data Integrity Check (Read Test)
Verify data integrity while the Drive Group (DG) is in a compromised state. This ensures data remains accessible and correct despite the drive failure.

sudo fio --filename=/dev/gdg0n1 \
  --ioengine=libaio --direct=1 --name=verify \
  --rw=read --bs=1048576 --numjobs=10 --iodepth=32 \
  --size=10% --offset_increment=10% \
  --cpus_allowed_policy=split \
  --verify=crc32c --verify_only=1 \
  --verify_state_load=1 --verify_fatal=1 --verify_dump=1 \
  --group_reporting=1

Recovery & Maintenance Operations

Step 8: Trigger Rebuild
Mark the physical drive as "Good" to initiate the rebuilding/recovering process.

sudo graidctl edit physical_drive 0 marker good

Step 9: Consistency Check
Once the Drive Group returns to OPTIMAL status, run a consistency check to ensure no data inconsistencies exist.

sudo graidctl start consistency_check manual_task -p stop_on_error
# Monitor status:
sudo graidctl describe consistency_check

Step 10: Copyback Operation
Execute a copyback task to restore data to the original slot (if applicable) or verify copyback functionality.
Note: Replace <SRC_PD_ID> and <DST_PD_ID> with the actual Source and Destination Physical Drive IDs.

sudo graidctl start copyback <SRC_PD_ID> <DST_PD_ID>
# Verify physical drive status:
sudo graidctl list physical_drive

Final Validation

Step 11: Post-Recovery Integrity Check
Perform a final read verification to ensure data integrity remains intact after the rebuild, consistency check, and copyback operations are complete.

sudo fio --filename=/dev/gdg0n1 \
  --ioengine=libaio --direct=1 --name=verify \
  --rw=read --bs=1048576 --numjobs=10 --iodepth=32 \
  --size=10% --offset_increment=10% \
  --cpus_allowed_policy=split \
  --verify=crc32c --verify_only=1 \
  --verify_state_load=1 --verify_fatal=1 --verify_dump=1 \
  --group_reporting=1