SSD

Beyond Spinning Disks

Pros

• Cheapest way to store large amounts of data

Cons

• Slowest component in most computers
• Fragile mechanical components can break
• Disk motor is extremely power hungry

---

Solid State Drives

NAND flash memory-based drives

• NOR: random access in Byte unit
  • HDD
• NAND: Access in Block, Page unit
  • ROM
  • Non-volatile memory

Props

• More resilient against physical damage
  • Disk가 없기 때문에 head를 읽거나 움직일 필요 x
• Reduced power consumption
  • No mechanical parts
• Much faster than hard drives
  • No penalty for random access
    • Each flash cell can be addressed directly
    • No need to rotate or seek
  • Extremely high throughput
    • Although each flash chip is slow, they are RAIDed

---

Challenges with Flash

Flash memory is written in pages, but erased in blocks

• Pages: 4-16KB, Blocks: 128-256KB
• Thus, flash memory can become fragmented (in block) -> Write amplication

Flash memory can only be written a fixed number of times

• Typically 3000 – 5000 cycles for MLC
• SSDs use wear leveling to evenly distribute writes across all flash cells

지울 때 저항이 발생하기 때문에 횟수 정해짐 -> 수준 맞추기

---

Write Amplification

• Once all pages have been written, valid pages must be consolidated
• A write triggers garbage collection / compaction
  • One or more blocks must be read, erased and rewritten before the write can proceed

Garbage Collection

• Vital for the performance of SSDs
• Many SSDs over-provision to help the GC
  • 240 GB SSDs actually have 256 GB
  • GC uses 16 GB
  • If use the all 240 GB, write on 16 GB and move later

---

The Ambiguity of Delete

• Goal : SSD wants to perform background GC
• Problem: most file systems don’t actually delete data
  • On linux, delete function is unlink() -> delete 표시만

Metadata is overwritten, but the file remains

1. File is written to SSD
2. File is deleted
3. GC executes
   • 9 pages look valid to the SSD

• Lack of explicit delete means the GC wastes effort copying useless pages (7 Files)
• Hard drives are not GCed, so this was never a problem

TRIM

• New SATA command TRIM
• Allow the OS to tell the SSD that specific LBAs are invalid
• Delete function makes files invalid

---

Wear Leveling

• each flash cell wears out after several thousand writes
• SSDs use wear leveling to spread writes across all cells
  -> 메모리 셀들의 마모를 균등하게 분산

Dynamic Wear Leveling

• Wait as long as possible before garbage collecting
• Blocks with pages that are not updated will not be rewritten
  • 업데이트 안 되는 블럭은 계속 쓰이지 않음

Static Wear Leveling

• Blocks with long lived data receive less wear
• SSD controller periodically swap long lived data to different blocks
  • block write 균형 맞춤
    
    

---

SSD Controllers

• SSDs are extremely complicated internally
• All operations handled by the SSD controller
  • Maps LBA to physical page
  • Keeps track of free pages, controls the GC
  • May implement background GC
  • Performs wear leveling

---

Flavors of NAND Flash Memory

Multi Level Cell

Consumer-grade drives

• Multiple bits per flash cell
• Higher capacity and cheaper than SLC flash
• Lower throughput due to the need for error correction(overhead)
• 3000-5000 write cycles
• Consumes more power

Single Level Cell

Expensive, enterprise drives

• One bit per flash cell
• Lower capacity and more expensive than MLC
• Higher throughput than MLC
• 10000 - 100000 write cycles