Measuring Performance

Question 1

Latency

Answer 1

time needed to complete a task. Usually related to execution time of a program.

Question 2

Throughput

Answer 2

no. of tasks completed in a fixed time. Related to bandwidth.

Question 3

Latency/CPU Performance equation/total time to run a program

Answer 3

time = dynamic instruction count x CPI x clock period

time = no. of instructions x CPI x clock period

Question 4

Dynamic instruction count

Answer 4

the number of instruction count at run time, related to execution of program. Different from static instruction count that does not change.

Question 5

Speedup

Answer 5

old time/new time
assuming old time is longer than new time

Question 6

Amdahl’s Law equation

Answer 6

speedup = 1/ ( (1 - f) + (f/s) )
f is the fraction of time spend in optimisable code
s is the speed up of the optimized code

Question 7

Amdahl’s Law in words

Answer 7

The maximum speedup of a program is limited by the fraction of the program that can not be improved.

Question 8

Lab 1 only uses functional execution driven. What does this mean?

Answer 8

The program is compiled with a certain optimization and the simulation runs the binary file.

Question 9

What did you do in Lab 1?

Answer 9

Wrote the simulator that simulates a different stage pipelines and verified its accuracy using a microbenchmark. Then ran the simulator with a given trace file to calculate the %drop in performance compared to an ideal pipeline with no stalls.

Question 10

-O0

Answer 10

no optimization, all data is saved on the stack

Question 11

-O1

Answer 11

optimization made, a lot of code