4. Performance Metrics and Perspectives

Question 1

System-Centric Performance Metrics

Answer 1

• Evaluate how well the overall system or any subset of its resources are being used, regardless of the quality of service

Question 2

Process-Centric Performance Metrics

Answer 2

• Evaluate how well a particular resource consumer is being served, regardless of the overall system resource efficiency

Question 3

Utilization

Answer 3

• The percentage of time that resource S was in use over time T
• Resource specific metric

Question 4

Throughput

Answer 4

• The number of requests that are processed by the system per unit time

Question 5

Effective Throughput (Goodput) (Bandwidth)

Answer 5

• The number of requests of value that are completed over a period T

Question 6

System Capacity

Answer 6

• The maximum throughput that a system can sustain

Question 7

Bottleneck

Answer 7

• The first resource to reach 100% utilization

Question 8

Response Time (Latency)

Answer 8

• The time elapsed from when the request is submitted until the service is finished

Question 9

Lag

Answer 9

• A measure of how much later after the deadline service was completed, on average

Question 10

Average Lag

Answer 10

• The mean lag over all tasks

Question 11

Average Lateness

Answer 11

• The mean lag for late tasks

Question 12

Starvation

Answer 12

• When a process is not served at all while others are continuously serviced

Question 13

Fairness

Answer 13

• For tasks with identical priority, a fair system will try to keep the ratio of response time and service time close to each other

Question 14

Speedup

Answer 14

T_old / T_new

Question 15

Slowdown

Answer 15

T_new / T_old

Question 16

Amdahl’s Law (Equation)

Answer 16

• If a resource is used while serving requests for a fraction of time f, making the resource x-times faster will improve overall performance by a factor of y
• y = 1 / (1 - f * (1 - 1/x))

Question 17

Amdahl’s Law (Qualitative)

Answer 17

• Speeding up the bottlenecked resource will result in the most improvement

Question 18

Amdahl’s Law (Parallelism)

Answer 18

• If 1/N of the total time taken by a program cannot be parallelized, the best you can achieve is to execute the program in 1/N of the original time

Question 19

Speedup (Parallelization)

Answer 19

• The ratio of the time it takes to execute the program with 1 processor and the time it takes to execute the program with j processors
• T(1) / T(j)

Question 20

Reliability

Answer 20

• The probability that the system functions correctly continuously for a specified period
• R(t) = e^(-t/MTBF)

Question 21

Availability

Answer 21

• The probability that a system is available (“up”) at an arbitrary point in time
• A = MTBF / (MTBF + MTTR)

Question 22

Mean Time Between Failures (MTBF)

Answer 22

• The average amount of time between a completed repair and the next failure
• MTBF = T_up / n_fail

Question 23

Mean Time To Repair (MTTR)

Answer 23

• The average amount of time for a failed system to be repaired
• MTTR = T_down / n_fail

Question 24

Predictability

Answer 24

• The difference between best/average-case response times and the worst-case response time