RAID 0 is fast—but it’s a data graveyard when a single disk fails.
In this post, we explore the real-world performance of 24 enterprise-grade SAS hard drives in a RAID 0 array, using the fio benchmark tool on a Dell PowerEdge R730xd server.

What is RAID?

RAID stands for Redundant Arrays of Independent Disks. It’s a data storage virtualisation technology that combines multiple physical drives into one logical unit to improve performance, reliability, or both.

Common RAID levels include:

• RAID 0 – Striping only. Maximises performance, but offers no redundancy. If a single disk fails, all data is lost.
• RAID 1 – Mirroring. Every disk holds the same data. Good redundancy, read performance improves, write performance stays similar to a single disk.
• RAID 5 – Striping with parity. Needs at least 3 disks. Offers fault tolerance with decent performance.
• RAID 6 – Similar to RAID 5, but with double parity. Can tolerate 2 disk failures. Requires at least 4 disks.
• RAID 10 – A combination of RAID 1 and RAID 0. Offers both high performance and redundancy.

This post focuses on RAID 0, which maximises speed but provides no data protection—hence the saying:

“RAID 0 is fast, but it’s a data crematorium.”

Test Setup

• Server: Dell PowerEdge R730xd
• RAID Controller: PERC H330 Mini (hardware RAID)
• Drives: 24 × 1.2TB 2.5" SAS enterprise hard drives
• Benchmark Tool: fio
• Operating System: Linux

The RAID 0 array was created using the megacli utility. The controller was configured with:

• Write policy: WriteThrough (writes bypass cache)
• Read policy: ReadAhead (improves sequential read speed)
• Direct I/O: Enabled (uses controller cache to improve read hits)

The total capacity shown was 26.2TB, not 28.8TB as one might expect from 24 × 1.2TB drives.
Why? Storage vendors use decimal units (1TB = 1000^4 bytes), while operating systems use binary units (1TB = 1024^4 bytes).
So:
1.2 × 1000⁴ ÷ 1024⁴ × 24 ≈ 26.2TB

Fio Test Script

 1
 2
 3
 4
 5
 6
 7
 8
 9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26

filename=fio.test  # originally /dev/sdb
size=100GB

# 4K random write
fio --iodepth=32 --numjobs=16 --size=$size --norandommap \
    --readwrite=randwrite --bs=4K --runtime=120 --filename=$filename \
    --ioengine=libaio --direct=1 --group_reporting --name=iops_write

sync && echo 1 > /proc/sys/vm/drop_caches

# 4K random read
fio --iodepth=32 --numjobs=16 --size=$size --norandommap \
    --readwrite=randread --bs=4K --runtime=120 --filename=$filename \
    --ioengine=libaio --direct=1 --group_reporting --name=iops_read

# Sequential write (4MB blocks)
fio --iodepth=32 --numjobs=1 --size=$size --norandommap \
    --readwrite=write --bs=4M --runtime=120 --filename=$filename \
    --ioengine=libaio --direct=1 --name=seq_write

sync && echo 1 > /proc/sys/vm/drop_caches

# Sequential read (4MB blocks)
fio --iodepth=32 --numjobs=1 --size=$size --norandommap \
    --readwrite=read --bs=4M --runtime=120 --filename=$filename \
    --ioengine=libaio --direct=1 --name=seq_read

Single Drive Benchmark

Each 1.2TB SAS drive was first tested individually for baseline performance:

• 4K Random Write: 343 IOPS

• 4K Random Read: 428 IOPS

• Sequential Write: 205.8 MB/s

• Sequential Read: 205.0 MB/s

24-Drive RAID 0 Benchmark

The full 24-disk RAID 0 array was then tested with the same parameters:

• 4K Random Write: 1192 IOPS

• 4K Random Read: 1236 IOPS

• Sequential Write: 3689.5 MB/s

• Sequential Read: 4644.3 MB/s

Performance Summary

Conclusion

This test demonstrates that RAID 0 can offer incredible throughput, especially for large, sequential operations. Read speeds above 4.6GB/s and write speeds above 3.6GB/s are impressive, even for 15K RPM enterprise drives.

However, the gains for random I/O are far less significant. More importantly, RAID 0 offers no data redundancy. If any one of the 24 disks fails, all data in the array is lost.

Recommendation: Use RAID 0 only for temporary data, caches, or situations where speed is paramount and loss is acceptable.

TL;DR

• RAID 0 with 24 SAS HDDs delivers massive sequential performance.
• Random IOPS scaling is limited by disk mechanics and controller architecture.
• Never use RAID 0 for critical data without proper backup.
• As the saying goes:

“RAID 0 is fun until someone loses a disk.”