11 - Performance & Optimisation

Question 1

What is implicit with the goal of parallelism?

Answer 1

Optimisation

Question 2

What are 2 ways to decide which parallelisation strategy is better?

Answer 2

Use theoretical measures

Measure the performance and compare

Question 3

What is theoretical performance?

Answer 3

Span/step and work complexity

Question 4

What is the most critical part when parallelising code?

Answer 4

The theoretical performance.

It can give you huge speedup gains

Question 5

What is latency?

Answer 5

The time it takes to complete a single task

Question 6

What is throughput?

Answer 6

The rate at which tasks can be complete

Question 7

What is better, higher or lower latency?

Answer 7

Lower

Question 8

What is better, higher or lower throughput?

Answer 8

Higher

Question 9

What does latency minimise?

Answer 9

Time at the expense of power

Question 10

What does throughput minimise?

Answer 10

Quantity of tasks processed per unit of time

Question 11

What is optimised for low latency computations, CPU or GPU?

Answer 11

CPU

Question 12

What is optimised for data-parallel and high throughput computations?

Answer 12

GPU

Question 13

What has the larger cache? CPU or GPU?

Answer 13

CPU

Question 14

What is speedup?

Answer 14

Compares the time T for solving the identical problem on one processor versus on p processors

Question 15

What is the ideal speedup?

Answer 15

Linear

Question 16

What is the formula for speedup?

Answer 16

S = Ts / Tp

Question 17

What is efficiency?

Answer 17

Measures the utilisation of hardware resources (return on hardware investment)

Question 18

What are 2 parallelisation metrics?

Answer 18

Speedup and efficiency

Question 19

What is the ideal efficiency?

Answer 19

1 (or 100%)

Question 20

What are 3 sources of performance loss?

Answer 20

Non-parallelisable computation (always small part that is serial)

Overhead (extra effort for communication between processors)

Under-utilisation (idle processors, slow memory)

Question 21

What is Amdahl’s Law?

Answer 21

The improvement to be gained from using a faster mode of execution is limited by the fraction of time that this mode is used

Question 22

What is the formula for Amdahl’s law?

Answer 22

s = 1/(1-f) + f/p

Question 23

What does Amdahl’s Law provide?

Answer 23

A theoretical upper limit on parallel speedup assuming that there are no costs for parallelism

Question 24

What overheads can be an extra cost of parallelisation?

Answer 24

Communication (shared memory)

Synchronisation (waiting for work to complete, barriers)

Computation (assignment of tasks to processors)

Memory requirements (parallel algorithms typical require more memory)

Question 25

What is under-utilisation?

Answer 25

Each hardware will have its maximum capabilities Not utilising hardware resources to their full potential decreases efficiency

Question 26

What is load imbalance?

Answer 26

Uneven distribution of work to processors

Question 27

What is power consumption proportional to?

Answer 27

The cube of processor frequency f

Question 28

What are 4 strategies to gain the best performance and optimisation?

Answer 28

Selection of the right algorithm Following basic principles for writing efficient code Architecture specific optimisation Micro-level optimisation

Question 29

What is (1-f), f and p in Amdahl's law?

Answer 29

``` (1-f) = fraction of serial code (0-1) f = fraction of parallel code (0-1) p = number of processors ```