Why CPU Access to Hard Drives Is So Slow: A Deep Dive into HDD, SSD, and I/O

Mechanical hard drives (HDD) and solid‑state drives (SSD) are the two most common external storage devices. Accessing data on them from the CPU is orders of magnitude slower than accessing cache or main memory. For example, a 4 KB random read from an SSD takes about 150 µs, while the same operation on an HDD requires roughly 10 ms, and a L1 cache reference is only 0.5 ns.

Because the CPU cannot read the disk directly, the operating system must first move data from the disk into main memory via an I/O operation. The article outlines three typical I/O methods:

I/O Operations

Programmed I/O – the CPU polls the device and writes each character (e.g., Hello World) itself, consuming the entire CPU core.

Interrupt‑driven I/O – the device raises an interrupt when it is ready, allowing the CPU to perform other work while the device handles data transfer.

Direct Memory Access (DMA) – a DMA controller transfers a whole buffer between memory and the device without CPU intervention, reducing interrupt overhead.

While DMA frees the CPU, it can still be slower than interrupt‑driven or programmed I/O if the controller cannot keep up with the device, causing the CPU to wait for DMA‑generated interrupts.

Mechanical Hard Drives

HDDs store data magnetically on rotating platters. A read or write requires the actuator arm to move the read/write head to the correct track (seek), the platter to rotate to the correct sector, and then the data to be transferred. This sequence typically costs about 10 ms, which dominates random I/O latency.

Because random I/O on HDDs is so expensive, database workloads that perform many random reads suffer severe performance penalties. Sequential reads are much faster (≈40 MB/s), highlighting the importance of minimizing random access.

Solid‑State Drives

SSDs use flash memory with no moving parts, allowing read/write operations to be performed electronically. Consequently, SSD latency is orders of magnitude lower than HDD latency, making them the preferred storage for performance‑critical servers and databases.

Summary

The CPU must go through I/O operations to reach data on external storage, incurring overhead from programmed I/O, interrupt‑driven I/O, or DMA. Mechanical HDDs add additional latency due to their moving‑part architecture, typically around 10 ms per random access, while SSDs provide much faster access. Although disks are not strictly required for booting (network or other media can be used), they remain the most common external storage in today’s computers.

CPU I/O overhead: each I/O method consumes CPU cycles and can become a bottleneck.

Mechanical HDD latency: queueing, seek, rotation, and transfer steps lead to ~10 ms per random I/O.