L14 - Storage / Memory

Question 1

List some different storage technologies

Answer 1

1. SRAM
2. DRAM
3. Flash

Question 2

List properties of SRAM.

Answer 2

1. Fast access
2. No refreshes to worry about
3. Simpler manufacturing
4. Less transistors
5. Higher cost

Question 3

List properties of DRAM.

Answer 3

1. Much higher capacity
2. More transistors
3. Lower cost

Question 4

What is SRAM suitable for?

Answer 4

registers and L1-L3 caches

Question 5

What is DRAM suitable for?

Answer 5

stand-alone memory chips

Question 6

What about embedded DRAM?

Answer 6

good for L3-L4 caches

Question 7

How is parallelism achieved in DRAM?

Answer 7

with multiple:
- Channels
- Ranks
- Banks
- Chips

Question 8

5 basic commands of DRAM

Answer 8

1. ACTIVATE (open a row)
2. READ (read a column)
3. WRITE
4. PRE-CHARGE (close row)
5. REFRESH

Question 9

DRAM Row Buffer Management Policies

Answer 9

1. open row policy
2. closed row policy
3. adaptive

Question 10

What are the pro and con of open row policy?

Answer 10

+ : Next access might need the same row -> row hit!
- : Next access might need a different row -> row conflict

Question 11

How is a DRAM address composed?

hint: 3 values

Answer 11

Address = (Bank x, Row y, Column z)

Question 12

How do the banks operate?

Answer 12

Banks operate independently, but share command/address/data pins.

• Each bank can have a different row active
• Can overlap ACTIVATE and PRE-CHARGE latencies (i.e. READ to bank 0 while ACTIVATING bank 1)

Question 13

What does a DRAM controller do?

Answer 13

1. map “Physical Address” to DRAM Address
2. buffer and schedule requests (to improve performance)
3. refresh: ensure correct operation
4. manage power consumption

Question 14

What is Flash?

Answer 14

semiconductor, non-volatile memory technology

Question 15

T/F: Disk is a magnetic storage technology.

Answer 15

T

Question 16

What are the characteristic of Flash storage compared to a mechanical disk?

Answer 16

1. lower latency
2. higher bandwidth
3. lower power
4. lighter weight, smaller size
5. shock resistance

Question 17

What are the characteristics of multi-level cell compared to single-level cell?

Answer 17

1. slower but denser
2. less reliable
3. cheaper

Question 18

Where is the multi-level cell technology used?

Answer 18

consumer drives (thumb, cheap SSDs)

Question 19

Where it the single-level cell technology used?

Answer 19

enterprise drives

Question 20

What are the two types of flash? What is the issue that they both have?

Answer 20

1. NOR
2. NAND

• both are damaged after some number of write/erase cycles

Question 21

How is parallelism achieved in flash storage systems?

Answer 21

With use of multiple
- controllers
- channels
- chips
- blocks
- pages

Question 22

What is the minimum unit of read/write?

Answer 22

page

Question 23

What is the minimum unit of erase?

Answer 23

block

Question 24

What kind of bit change happen during a write?

Answer 24

Only 1 -> 0 transitions are allowed

Question 25

What happens to the bits in the block during erase?

Answer 25

All bits are set to 1

Question 26

Why does 99% read workload get half of the throughput compared to 100% read workload?

Answer 26

Because read waits for slow writes/erases to complete and each chip can process one operation at a time

Question 27

What if we want to share Flash (use same device for multiple applications, different tenants)?

Answer 27

Unpredictable performance: reads of one application suffer from interference with write of other application

Question 28

How would one circumvent that issue?

Answer 28

isolation mechanisms: 1. I/O scheduler 2. rate limit

Question 29

List 3 reliability issues of Flash.

Answer 29

1. Wear out 2. Writing disturb 3. Read disturb

Question 30

What does wear out mean?

Answer 30

Flash cells are physically damaged programming and erasing them.

Question 31

What does writing disturb mean?

Answer 31

Programming pages can corrupt the values of other pages in the block.

Question 32

What does read disturb mean?

Answer 32

Reading data can corrupt the data in the block. It takes many reads to see this effect.

Question 33

What is FTL?

Answer 33

Flash translation layer: tries to make flash look like a disk 1. Exposes a block-based interface (like a disk) 2. Manages program/erase granularity mismatch 3. Equalises wear 4. Delivers high performance

Question 34

What kind of address mapping is FTL responsible for?

Answer 34

logical --> physical

Question 35

What is write amplification?

Answer 35

number of physical pages written to flash for every logical page (ratio is above 1)

Question 36

How can we reduce write amplification?

Answer 36

1. reserve extra capacity on Flash (more clean blocks will be available, garbage collect less often) 2. make your program write sequentially 3. store frequently written data in memory

Question 37

How does one implement wear leveling?

Answer 37

1. Keep statistics on number of write/erase cycles 2. Erase blocks with low erase counts, move data to new location when current block starts to wear out

Question 38

How/why do FTL implementations use DRAM?

Answer 38

1. Store logical to physical mappings for fast lookups 2. Queue requests 3. Buffer writes to hide high write latency

Question 39

List Flash SSD server interfaces?

Answer 39

1. SATA 2. SAS 3. PCIe

Question 40

What happens if your PC fails due to an error in the SSD and reboots?

Answer 40

The garbage will still be there. :(