Lecture 07

Question 1

What is virtual memory?

Answer 1

When two or more levels of memory hierarchy are made to look like one.
You get the speed of the top level, and the size of the bottom level.

Question 2

What decade was virtual memory invented in?

Answer 2

The 1960s

Question 3

What positive effect does virtual memory have on multi-processing?

Answer 3

Each process can be given its own virtual address space, which means a stray pointer in one process cannot harm another one.

Question 4

When caching, what is a line?

Answer 4

A chunk that data has been divided into.

Question 5

When caching, what are addresses split into?

Answer 5

A chunk identifier and an offset within the chunk.

Question 6

What are the main steps involving the cache to perform when a user accesses data?

Answer 6

• Look up the chunk identifier to see if it’s in fast memory, and where
• If not:
  
  • Find or make space in the fast memory
  • Transfer the chunk from slow memory

Question 7

What is ‘Spatial Locality’?

Answer 7

Memory being accessed is often near memory accessed recently.
Chunk size is often chosen to exploit this.

Question 8

What is ‘Temporal Locality’?

Answer 8

Accesses to the same area of memory tend to be grouped in time.

Question 9

In a Direct Mapped Cache, how many different places could a given chunk be in?

Answer 9

1.

Multiple chunks may map to the same place, however.

Question 10

In a Fully Associative Cache, how many different places can a given chunk be in?

Answer 10

A chunk in a FAC can be placed anywhere.

Question 11

What is the typical line length of a Fully Associative Cache?

Answer 11

4 KiB

They can, however, be much bigger.

Question 12

What do the least significant 12 bits represent in the address of a line of a Fully Associative Cache?

Answer 12

The chunk’s offset within the line.

Question 13

How does a Fully Associative Cache know if a chunk is in fast memory?

Answer 13

It compares a chunk’s tag and offset against a lookup table which stores everything that’s in the cache.

Question 14

What is it called when a cache gets rid of something?

Answer 14

Evict

Question 15

What are the three sections that a Direct Mapped Cache address is split into?
What is each section used for?

Answer 15

Offset - Length determines the length of a line
Index - Length determines the number of lines in a cache
Tag - Whatever remains

Question 16

If a Direct mapped Cache has a 5 bit offset, how long is each line?

Answer 16

2^5 = 32B

Question 17

If a Direct Mapped Cache has a 17 bit index, how many lines are there in the cache?

Answer 17

2^17 = 131072 lines

Question 18

If a Direct Mapped Cache has a 17 bit index, 5 bit offset, and 10 bit tag, how much memory is needed for the cache?

Answer 18

2^17 * (2^5 + 10 + 1) ~= 5.4MiB

Question 19

If a Direct Mapped Cache has a X bit index, Y bit offset, and Z bit tag, what is the formula to work out the memory needed for the cache?

Answer 19

2^X * (2^Y + Z + 1)

The 1 is for the valid bit (whether the line is in use or not)

Question 20

What are the steps a Direct Mapped Cache follows when looking up an address?

Answer 20

1. Use index to find the line in the cache
2. If the valid bit is off, the line is not in the cache (miss)
3. Otherwise, compare tags.
4. If the tag is the same, then the line is in the cache (hit)
5. Otherwise, ‘wrong’ line in cache (miss) and will need to be replaced by the ‘correct’ line (evict)

Question 21

What makes a Direct Mapped Cache wasteful?

Answer 21

Two lines with the same index can compete for the same line in the cache, while there’s lots of free space elsewhere in the cache.

Question 22

Does Direct Mapped Caching make a choice about whether or not to evict a line?

Answer 22

No.

The older line is always replaced by the newer one.

Question 23

In a 4-way associative cache, how many lines are there per index?

Answer 23

4.

For an N-way associative cache, each index has N lines.

Question 24

In an 8-way associative cache, how many lines does it check when looking for a specific line.

Answer 24

8.

An N-way cache with look at all associated lines in an index.

Question 25

How does an Associative Cache choose which line to evict on a ‘miss’?

Answer 25

It evicts the line that was used longest ago.

Question 26

What is ‘write-through’ in terms of cache operation?

Answer 26

Whenever something in the cache is written to, it also writes to slow memory straight away.
This keeps both copies up to date.

Question 27

What is ‘write-back’ in terms of cache operation?

Answer 27

The cache does not write the data back to slow memory until a ‘later time’. This can be as long as until eviction.

Question 28

What are the 9 levels of the memory hierarchy?

Answer 28

• Regs
• L1 Cache
• L2 Cache
• L3 Cache
• DRAM
• Flash
• Local DIsk
• Network Disk
• Cloud, Tapes

Question 29

Memory near the top of the hierarchy gets which of the following:

• More expensive
• Larger
• Faster

Answer 29

More expensive and faster.

It also becomes smaller.