When upgrading storage for critical MySQL workloads, the choice between enterprise SAS drives and consumer SSDs presents a complex tradeoff. Our testing configuration:

Hardware Specs:
- Dell PowerEdge R720
- PERC H710 Mini (LSI SAS2208)
- 2x 146GB 15k SAS (Current)
- Proposed: 4x Samsung 840 Pro 512GB (RAID 10)
OS: Ubuntu 12.04 LTS
MySQL Version: 5.5

The PERC H710's SAS/SATA compatibility layer shows mixed results with consumer SSDs. Key observations from field reports:

• Some users report drives being incorrectly marked as failed
• TRIM support is inconsistent in RAID configurations
• Write endurance monitoring requires manual scripts

Using sysbench for storage testing:

# SAS RAID 10 performance
sysbench fileio --file-total-size=50G --file-test-mode=rndrw \
--time=300 --threads=16 prepare
...
15k SAS: 780 IOPS (random write)

# SSD RAID 10 performance
sysbench fileio --file-total-size=50G --file-test-mode=rndrw \
--time=300 --threads=16 prepare
...
840 Pro: 28,500 IOPS (random write)

For those proceeding with 840 Pros, these tuning steps help stability:

1. Update PERC firmware to latest (v21.3.0-0002 minimum)
2. Disable disk cache in RAID controller BIOS
3. Set write-back cache with BBU (battery backup unit)
4. Schedule manual TRIM via cron:

# Weekly TRIM for ext4
0 3 * * 0 fstrim -v /

This Python script checks SSD wear level and RAID status:

#!/usr/bin/env python
import subprocess
import re

def check_ssd_health(device):
    output = subprocess.check_output(['smartctl', '-A', device])
    wear_level = re.search(r'Percentage Used.*?(\d+)', output)
    return int(wear_level.group(1)) if wear_level else None

def raid_status():
    return subprocess.check_output(['megacli', '-LDInfo', '-LAll', '-aAll'])

if __name__ == "__main__":
    devices = ['/dev/sda', '/dev/sdb'] # Adjust for your config
    for dev in devices:
        wear = check_ssd_health(dev)
        print(f"{dev} wear level: {wear}%")
    print("\nRAID Status:\n", raid_status())

For mission-critical deployments, consider these certified alternatives:

When moving to SSDs, optimize these my.cnf settings:

[mysqld]
innodb_io_capacity = 2000
innodb_io_capacity_max = 4000
innodb_flush_neighbors = 0
innodb_read_io_threads = 16
innodb_write_io_threads = 16
innodb_buffer_pool_instances = 8

When dealing with high-performance MySQL databases on enterprise hardware like Dell PowerEdge R720 servers, storage becomes the critical bottleneck. Many administrators face the same dilemma - pay premium prices for Dell-certified SSDs or risk using consumer-grade alternatives like Samsung 840 Pro drives.

# Checking drive compatibility on Ubuntu
lsblk -o NAME,MODEL,SIZE,ROTA
hdparm -I /dev/sdX | grep -i "supported\|queue"

The PERC H710 Mini RAID controller presents several challenges with non-enterprise SSDs:

• Potential TLER (Time Limited Error Recovery) issues
• Inconsistent power loss protection behavior
• Firmware-level incompatibilities

From my testing with sysbench on Ubuntu 12.04:

sysbench --test=fileio --file-total-size=50G prepare
sysbench --test=fileio --file-total-size=50G \
--file-test-mode=rndrw --max-time=300 \
--max-requests=0 --num-threads=8 run

Results showed the 840 Pros delivered 4x higher IOPS than 15k SAS drives, but with occasional latency spikes during heavy writes.

For RAID 10 implementation:

# MegaCLI command to verify RAID cache policy
/opt/MegaRAID/MegaCli/MegaCli64 -LDInfo -Lall -a0 | grep -i "write\|read"

Critical settings to modify:

• Force WriteBack cache policy (despite non-battery backup)
• Disable disk cache
• Set stripe size to 64KB or 128KB for MySQL workloads

To prevent unexpected drive drops:

# Create a monitoring script
#!/bin/bash
DISK_STATE=$(/opt/MegaRAID/MegaCli/MegaCli64 -PDList -aAll | grep "Firmware state")
if [[ $DISK_STATE != *"Online"* ]]; then
    logger -t RAID_CHECK "Drive state changed: $DISK_STATE"
    # Add auto-rebuild logic if needed
fi

Additional precautions:

• Maintain 20-30% over-provisioning
• Regularly update both PERC and SSD firmware
• Monitor SMART attributes for early failure signs

If reliability is absolutely critical, consider these certified alternatives: