Intro

Question 1

weak strong scaling

Answer 1

weak: problem size per HW thread/core etc stays the constant
strong: general size stays constant

Question 2

amdahl law

Answer 2

(i-f)T + fT/p

Question 3

parllel eficiency

Answer 3

speedup/p

Question 4

what can limit parallelization

Answer 4

1. comm and synch
2. resource bottlenecks
3. network conflicts
4. multiple processes on same node
5. load imbalance

Question 5

superlinear speedup

Answer 5

more bandwisth and fit in cache

Question 6

data and function parallelism

Answer 6

data: SPMD
less overhead, but load imbalance

functional: pipelining
good load balance, but overhead