SSD Architecture and Performance

Understanding SSD Architecture and Performance

Solid‑State Drives (SSDs) have revolutionized storage by replacing spinning magnetic disks with NAND flash memory. While they deliver impressive read speeds, their internal architecture introduces unique challenges that affect write performance, endurance, and overall reliability. This course explains the core concepts behind SSD operation, including page‑level programming, block erasure, write amplification, wear leveling, and garbage collection. Mastering these topics helps you optimize system design, troubleshoot performance bottlenecks, and choose the right SSD for your workload.

Page, Block, and Erase Fundamentals

What Happens When a Page Becomes Dirty?

In NAND flash, data is written to pages (typically 4 KB to 16 KB). However, pages cannot be overwritten directly. When a page is modified, it becomes dirty and must be treated as invalid. The SSD controller cannot simply erase a single page; instead, it must erase the entire block that contains that page.

• Each block consists of multiple pages (e.g., 5 pages = 20 KB).
• Erasing a block resets all bits to 1, making every page inside the block ready for new data.
• Because erasure is performed at block granularity, the controller must move any still‑valid pages elsewhere before erasing.

Therefore, the correct answer to the quiz question is: The entire block containing the dirty page must be erased.

Write Amplification: The Hidden Performance Killer

Why Does Write Amplification Degrade SSD Performance Over Time?

Write amplification occurs when the amount of data physically written to NAND exceeds the amount of data the host requested. This extra work is mainly caused by garbage collection and wear‑leveling activities that relocate valid pages before a block can be erased.

• More data movement means more program/erase (P/E) cycles.
• Each additional internal write consumes bandwidth that could otherwise serve host writes.
• As the drive fills, the controller spends increasing time consolidating data, so the effective write speed drops.

The quiz answer confirms this: Because extra data must be moved and rewritten during garbage collection.

SEO tip: Include keywords such as "SSD write amplification", "garbage collection impact", and "SSD performance degradation" to improve discoverability.

Wear Leveling: Extending SSD Lifespan

Primary Purpose of Wear Leveling

Each NAND cell can endure only a limited number of erase cycles (typically 1,000–10,000). Wear leveling spreads these cycles evenly across the entire flash array, preventing premature wear of hot‑spot blocks.

• Dynamic wear leveling moves newly written data to the least‑used blocks.
• Static wear leveling periodically relocates long‑lived (static) data from low‑wear blocks to higher‑wear ones.

Thus, the correct answer is: To distribute erase/write cycles evenly across all blocks.

Practical Example: Calculating Pages Required for a Write

Scenario

An SSD uses a 4 KB page size and a block contains 5 pages (20 KB total). Two pages are already occupied, and a 12 KB file needs to be written.

Because 12 KB equals three 4 KB pages, the SSD can place the new data into the three free pages of the same block. No extra pages are required for metadata in this simplified model.

Answer: Three pages, because the block can hold up to five pages.

Key take‑aways (as highlighted in the original explanation):

• Page size = 4 KB.
• Block = 5 pages = 20 KB.
• 12 KB file = 3 pages.
• Two pages already used → three free pages available.

Mnemonic: “4 KB × 3 = 12 KB → 3 pages free”. Visualize the block as a box with five compartments; two are filled, three remain empty for the new data.

TRIM: Communicating Deletions to the SSD

Which Management Technique Directly Addresses Undeleted Data?

Most traditional file systems mark data as deleted but do not actually erase it on the storage medium. The TRIM command informs the SSD which logical block addresses (LBAs) are no longer needed, allowing the controller to treat those pages as invalid and reclaim them during garbage collection.

• TRIM reduces unnecessary write amplification.
• Operating systems such as Windows, Linux, and macOS issue TRIM automatically for SSD‑aware file systems.

Correct answer: The TRIM command that informs the SSD of invalid LBAs.

Mapping Tables and DRAM Constraints

Why Can’t Large SSDs Use One‑to‑One Logical‑Physical Mapping?

In a simple one‑to‑one mapping, the controller would need a table entry for every physical page. As capacities grow (e.g., 1 TB with 4 KB pages → 256 million entries), the table would consume hundreds of megabytes of DRAM, far exceeding typical controller memory budgets.

• Large mapping tables increase latency and power consumption.
• To stay within DRAM limits, SSDs employ hierarchical or hybrid mapping schemes (e.g., block‑level, page‑level, or log‑structured).

Thus, the correct answer is: The mapping table would require too much DRAM memory.

Mnemonic: “Grande Capacità = Grande Tabella = Grande Problema” (GC‑GT‑GP). Imagine a phone book that becomes too thick to fit in your pocket—similarly, the mapping table can outgrow the controller’s DRAM.

Garbage Collection Strategies

Selecting a Block Containing Only Dead Pages

During garbage collection, the controller aims to reclaim space with minimal data movement. Choosing a block where all pages are already invalid (dead) eliminates the need to copy any live data before erasing.

• No extra read‑modify‑write cycles are required.
• Erase latency is reduced, freeing the block for future writes faster.

Correct answer: No live data needs to be copied before erasing the block.

Write Speed vs. Write Amplification

How Does Amplification Influence Real‑World Throughput?

Effective write speed is the host‑requested throughput divided by the write amplification factor. As amplification rises, the SSD must perform more internal writes per host write, lowering the observable speed.

• Example: If the host asks for 200 MB/s and the amplification is 2×, the SSD actually writes 400 MB of NAND data, halving the effective speed.
• Manufacturers mitigate this by over‑provisioning, optimizing garbage collection, and using advanced wear‑leveling algorithms.

Quiz answer: Effective write speed drops as write amplification increases.

Mnemonic: “A‑more A‑more S‑slow” – more Amplification, slower Writes.

Key Takeaways and Best Practices

• Never overwrite a page directly; always erase the whole block.
• Write amplification is the primary factor that reduces SSD write performance over time.
• Wear leveling spreads P/E cycles, extending drive lifespan.
• TRIM is essential for modern operating systems to keep SSDs efficient.
• Mapping tables must be sized to fit within the controller’s DRAM, prompting hierarchical mapping schemes.
• Garbage collection prefers blocks with only invalid pages to minimize data movement.
• Higher write amplification directly translates to lower effective write speeds.

By understanding these mechanisms, you can make informed decisions about SSD selection, firmware tuning, and system configuration to achieve optimal performance and durability.