Which storage option for MySQL data directory typically offers the worst performance in a highly concurrent, OLTP-heavy, IO-bound workload?

Question

Options

• Abattery-backed locally-attached RAID 5 array
• BiSCSI Lun
• CSAN (Fibre Channel) Lun
• DNFS (Networked File System) mount

• A
  58% (33)
• B
  23% (13)
• C
  5% (3)
• D
  14% (8)

Explanation

RAID 5's write penalty makes it the worst choice for this workload. Every single write operation requires four physical I/Os: read old data, read old parity, write new data, write new parity. Under highly concurrent OLTP - where dozens of small random writes are hitting the array simultaneously - this 4x write amplification compounds into a severe bottleneck, even with a battery-backed controller cache (which only defers the problem until the cache saturates).

Why the distractors are wrong:

• B (iSCSI): Block-level protocol with some network latency overhead, but no write penalty - each write is still one logical write to the LUN.
• C (SAN/FC): Fibre Channel is explicitly engineered for high-performance, low-latency database workloads; it's typically the best option here.
• D (NFS): File-level and not ideal for MySQL data files, but modern NFS with directio avoids double-buffering and carries no write-amplification penalty like RAID 5 does.

Memory tip: Think "RAID 5 = 4 I/Os per write." In an IO-bound workload, any storage technology that multiplies your I/O count is catastrophic - and RAID 5 does exactly that on every write. The network-attached options add latency; RAID 5 adds volume.

No community discussion yet for this question.