Speedup Amdahl Gustafson Flashcards

1
Q

What is scalability in prallel programming?

A
  • speedup when we increase number of processors
  • what happens if we have infite processors
  • linear speed up is the wish
How well did you know this?
1
Not at all
2
3
4
5
Perfectly
2
Q

Define parallel Speedup.

A

Speedup Sp on p CPU’S –> Sp = T1/Tp

(informally: time on one processor / time on p processors)

How well did you know this?
1
Not at all
2
3
4
5
Perfectly
3
Q

Define Efficiency with Sp and p.

A

Efficiency: Sp / p

(informally: Speedup / number of processors)

How well did you know this?
1
Not at all
2
3
4
5
Perfectly
4
Q

Why is Sp < p in the normal case?
Means that the speedup is not linear to the amount of processors.

A
  • Programs may not contain enough parallelism
  • overheads introduced by parallelization
  • architectural limitations
How well did you know this?
1
Not at all
2
3
4
5
Perfectly
5
Q

Assume Parallel program:

  • sequential part: 20%
  • parallel part: 80% (assume it scales linearly)
  • T1 = 10
A

see picture

How well did you know this?
1
Not at all
2
3
4
5
Perfectly
6
Q

Formula for Efficiency

A

Ep = Sp / p

How well did you know this?
1
Not at all
2
3
4
5
Perfectly
7
Q

Formula Speedup

A

Sp = T1 / Tp

How well did you know this?
1
Not at all
2
3
4
5
Perfectly
8
Q

Gives the ingrediants to Amdahl’s Law. First define Sp without using f. Thand use f as non parallizible part.

A

Amdahls law gives a bound on speedup.

Speedup is Sp := T1 / Tp

T1 := Wser + Wpar (W for work)
Tp ≥ Wser + (Wpar / P) (P number of processors)

Hence we have:
Sp ≤ (Wser + Wpar) / (Wser + (Wpar / P))

NOW using f for nonparallizible part:

Wser = f\*T1
Wpar = (1-f)\*T1

Sp ≤ 1 / (f + ( (1-f) / P)

How well did you know this?
1
Not at all
2
3
4
5
Perfectly
9
Q

Give Amdahls Law formula.
First for Wser than for Wpar and than for Sp using f.

A
Wser = f\*T1
Wpar = (1-f)\*T1

Sp ≤ 1 / (f + ( (1-f) / P)

How well did you know this?
1
Not at all
2
3
4
5
Perfectly
10
Q

What is a takeaway of Amdahls Law

A

all non- parallel parts of a program (no matter how small) can cause problems for scalability

How well did you know this?
1
Not at all
2
3
4
5
Perfectly
11
Q

What is the difference to Amdahls Law using Gustafsons Law? IMPORTANT

A

The time stays the same when thinking about Gustafsons Law.

Even if more processors are used the runtime stays the same.
More processors allow to solve larger problems in the same time.

How well did you know this?
1
Not at all
2
3
4
5
Perfectly
12
Q

What is the formula to Gustafsons Law. Give Sp

A

f := sequential part

Sp = f + p( 1 - f) = p - f ( p - 1 )

How well did you know this?
1
Not at all
2
3
4
5
Perfectly
13
Q

Which law is used when talking about the pessimistic case?

A

Admhahl’s Law

How well did you know this?
1
Not at all
2
3
4
5
Perfectly
14
Q

Which law is used when talking about the optimistic case?

A

Gustafson’s Law

How well did you know this?
1
Not at all
2
3
4
5
Perfectly
15
Q

Which law is used when talking about constant work?

A

Amdahl’s Law

How well did you know this?
1
Not at all
2
3
4
5
Perfectly
16
Q

Which law is used when talking about constant time?

A

Gustafson’s Law

17
Q

Which law is used when talking about the worst case?

A

Amdahl’s Law

18
Q

With wich law is linear speedup possible only theoretically? And why?

A

With Gustafson’s Law since the time to perform the taks stays the same using more processors