Performance (Computing)

Perfomance measurement: Execution Time and Throuput/Bandwidth. (Decreasing execution time always improves thruput) And intuitively performance is inversly related to execution time: $\text{Performance}=\frac{1}{\text{Execution Time}}$.

Measuring Performance §

• Program Execution Time (is the sum of…)
  • CPU Time (e.g. 10 picoseconds)
    • User CPU Time
    • System CPU Time
  • I/O Time
• Clock Cycle (= tick, clocks, cycles) e.g. one tick on this CPU is 250 picoseconds.
• Instruction Count: Number of instructions in a program.

⇒ Clock Per Instruction (CPI) / IPC (Instructions Per Clock)

Thus

$$\begin{gathered} \text{CPU Time (s)}=\text{\# of cycles}\times \text{Tick Time (s)} \\ \text{\# of cycles}=\text{\# of Instructions}\times \text{avg. CPI} \end{gathered}$$

And:

$$\text{CPU Time (s)}=\frac{\text{\# of Instructions}\times \text{CPI}}{\text{Clock Rate (Hz)}}$$

The Three Factors of Performance: Clock Rate / Instruction Count / CPI

$$
\text{Performance}= \frac{\text{Seconds}}{\text{Program}}\\=\frac{\text{Instructions}}{\text{Program}}\times \frac{\text{Clock Cycles}}{\text{Instruction}}\times \frac{\text{Seconds}}{\text{Cycle}}\\\\
=\text{Instruction Count} \times \text{CPI} \times \text{Clock Rate}


$$

List of Common Misconcpetions Performance §

Improving one aspect doesn’t improve overall performance porportionally

Amdahl’s Law:

$${\text{Execution time}}_{\text{after improvment}}=\frac{{\text{Execution time}}_{\text{affected by improvement}}}{\text{Amount of improvment}}\times {\text{Execution time}}_{\text{after improvment}}$$

→Some perfomance targets are impossible with partial improvements. See example on p.50.