Storage I/O Performance, RAID Penalties, and Bandwidth Calculation Guide

In performance‑optimization projects, it is essential to align storage configurations with real business requirements, analyzing end‑to‑end components such as host ports, storage subsystems, and backend disks.

IO Aggregation and Full‑Stripe Writes – When I/O is aggregated to a full‑stripe size, pre‑read is unnecessary and RAID write penalties are avoided; for example, RAID5‑5 small writes require two pre‑reads and one parity write, expanding one I/O into four, whereas full‑stripe writes expand four data I/Os into five, improving efficiency.

Storage IO Merge Capability – The ability to merge I/O varies by storage vendor and workload. Random I/O typical of database workloads rarely achieves full‑stripe merging, while sequential small I/O can often be merged into larger I/O.

Business I/O Models – Four common models are described: OLTP (small random I/O, 8 KB, read/write ratio ~3:2), OLAP (large sequential I/O, 200 KB+, read‑heavy), VDI (mixed, startup/login storms, latency ~10 ms), and SPC‑1 (industry‑standard random I/O benchmark, 4 KB, read/write ratio ~4:6).

SSD, SAS, NL‑SAS Comparison – Performance characteristics and advantages of each drive type are outlined, emphasizing SSD’s superiority for random small I/O.

FC Link Bandwidth Calculation – The theoretical bandwidth of an 8 Gbps FC link is computed as 787.5 MB/s using the formula: link clock × encoding efficiency × protocol efficiency ÷ 8 ÷ 1024 ÷ 1024. Real‑world bandwidth is lower due to protocol overhead and hardware limitations.

Impact of Parity Writes – In RAID5/6, each data write may trigger additional parity I/O, reducing usable bandwidth; for sequential full‑stripe writes, parity overhead is minimized.

Read/Write Ratio Influence – Higher write ratios increase storage resource consumption, especially under RAID5/6 where write penalties amplify I/O count.

RAID Level Performance – RAID10, RAID5, and RAID6 are compared, showing that write‑heavy workloads suffer more on RAID5/6 due to parity overhead, while RAID10 offers better performance at the cost of capacity.

Sequential vs Random I/O – Sequential I/O benefits from lower seek times and higher cache pre‑fetch efficiency, whereas random I/O incurs higher latency, especially on mechanical disks.

IO Size Effects – Small I/O (≤16 KB) is measured by IOPS, large I/O (≥32 KB) by bandwidth; larger I/O reduces IOPS but increases throughput, influencing drive selection and RAID choice.

Cache Acceleration – Cache improves write performance via write‑back, write‑hit, and write‑merge mechanisms, and enhances read performance through cache hits and pre‑fetch, with “full‑hit” scenarios delivering the highest IOPS.