Computer Architecture Basics and Performance

Introduction to Architecture

• Instruction Set Architecture (ISA): An assembly language programmer's view of the microprocessor.
• Assembler: Translates symbolic assembly language code into binary machine language code.

Computer Performance Metrics

• Execution Time (Response Time): Time to complete a program or task.
• Throughput (Bandwidth): Number of programs or tasks completed per time unit.
• Execution Time (t_exec, i.e., elapsed time, wall time, real time, response time): Elapsed seconds from start to finish. Includes:
  • Executing instructions
  • Disk access
  • Memory access
  • I/O activities
  • OS overhead
• User CPU Time (t_cpu(user)): Seconds the CPU spends executing the program's instructions.
• System CPU Time (t_cpu(sys)): Seconds the CPU spends executing OS (kernel) instructions on behalf of the program.
• System Performance: Execution time on an unloaded single-tasking system. perf_sys = 1 / t_exec. Includes OS overhead, so system performance is highly dependent on the efficiency of the OS.
• CPU Performance: The inverse of user CPU time. perf_cpu(user) = 1 / t_cpu(user). Excludes OS overhead, so CPU performance is highly dependent on the CPU organization. To increase CPU performance, user CPU time must decrease.
• Bottleneck: Something which restricts performance. The canonical computer system bottleneck is CPU to main memory access time.
• To maximize performance on program P on computer system X, minimize execution time of P on X: perf_X(P) = 1 / t_exec,X(P)
• Relative Performance (n) of computer X compared to computer Y (on program P): perf_{rel,X(P),Y(P)} = perf_X(P) / perf_Y(P)

Time and Frequency Units

• Time Units (Period):
  • Millisec (ms) = 10^-3 sec
  • Microsec (μs) = 10^-6 sec
  • Nanosec (ns) = 10^-9 sec
  • Picosec (ps) = 10^-12 sec
• Frequency Units:
  • Hertz (Hz) = 1
  • Kilohertz (KHz) = 10³
  • Megahertz (MHz) = 10⁶
  • Gigahertz (GHz) = 10⁹ times per second

System Clock and CPU Time

• System Clock: Device which generates the master clock signal. Coordinates activities of the system. Clock frequency (clock rate, tick rate) and clock period (clock cycle time, cycle time) are inversely related: f_clock = 1 / p_clock.
• Relationship between CPU performance (user CPU time) and system clock: t_cpu(user) = prog clocks * p_clock; t_cpu(user) = prog clocks / f_clock. To decrease user CPU time: decrease program clocks, and/or decrease clock period (increase clock frequency).
• Static Instruction Count (icount_stat): Number of machine language instructions in the program.
• Dynamic Instruction Count (icount_dyn): Number of machine language instructions that are executed.