CIS 273 Computer Organization

1
CIS 273Computer Organization Design

• Instructor Dr. Michael Geiger
• Spring 2007
• Lecture 6 Computer Performance

2
Lecture outline

• Exam notes stats, solution, regrades
• Computer performance
• Performance metrics
• Execution time
• Improving performance
• Tradeoffs

3
Exam statistics

• Mean 68.9
• Median 71.3
• Std. deviation 14.5
• High 91

• Avg. by question
• 9 / 18 (50)
• 10 / 13 (77)
• 7.4 / 12 (61)
• 12.6 / 16 (79)
• 14 / 20 (70)
• 9 / 11 (83)
• 6.7 / 10 (67)

4
Regrade requests

• Submit IN WRITING by end of class next Tuesday
  (4/3)
• Must explain why regrade is necessary
• Bad example I deserve more points on question
  5.
• Good example On question 5, part III, I used
  the refined hardware multiplication method, not
  the pencil and paper method. My answer differs
  from the one in the solution, but I think its
  still correct.

5
Performance

• Measure of how well a computer performs a users
  applications
• Why study performance?
• As a user/buyer, you can make better purchasing
  decisions
• As an engineer, you can make better design
  decisions
• Measuring performance requires basis for
  comparison and metric for evaluation
• Analogy if youre buying a car, what do you look
  for?
• Speed (Maximum? Acceleration?)
• Engine performance (Horsepower?)
• MPG
• Cost
• Some combination of the above (e.g.,
  price/performance)

6
Latency vs. throughput

• Two metrics for measuring performance
• Latency How long does it take to complete a
  particular task?
• Also called execution time or running time
• Usually measured in time (e.g., microseconds)
• Throughput How many tasks can be performed in a
  unit of time?
• Usually measured in units per time (e.g.,
  megabytes/second)
• Example We want to transport people between
  Providence and Boston, using either
• A 2-seat BMW _at_ 150 MPH, or
• A chartered bus (45 seats) _at_ 60 MPH
• Which is better?

7
Time-based performance metric

• Choose performance measures that track running
  time
• Higher performance ? less time to run the
  application
• Execution time on a computer is typically divided
  into
• User time Time spent executing instructions in
  the user code
• System time Time spent executing instructions in
  the OS on behalf of the user code (e.g., opening
  files)
• Other Time when the system is idle or executing
  other programs

8
Illustration of execution time

• Real or wall clock time is the sum of all
  three.
• Note File access delays sometimes counted as
  idle even though theyre yours.
• Normally execution time is measured as
• user time system time

9
Measuring time

• Clock ticks refer to a full clock cycle
• Cycle time (period) time between ticks
  seconds per cycle.
• Clock rate (frequency) cycles per second
• 1 Hz 1 cycle/sec
• Example a 200MHz clock has a cycle time of

10
Improving performance

• Improving performance means decreasing the time
  to run the program
• So to increase performance
• Decrease the number of cycles required for a
  program, or
• Decrease the cycle time (i.e., increase the clock
  frequency)
• A program requires a set number of instructions
  (for a given ISA)
• How do we relate cycles and instructions?

11
Cycles and instructions

• In most processors
• A single instruction (one line of assembly code)
  requires more than one clock cycle
• Different (types of) instructions take different
  numbers of cycles (e.g., integer vs. floating
  point)
• Another metric Cycles per instruction (CPI)
• The smaller, the better
• The average CPI of a program depends on how many
  of each instruction type is executed
• Clock cycles for a program
• Instructions executed Average CPI

12
Example Execution time

• A program requires executing 100 million
  instructions on a processor which typically takes
  2 CPI with a 2GHz clock. How much time will the
  program take?
• Answer

13
Performance comparison

• If machine X is n times faster than machine Y
• Example
• Time of Concorde vs. Boeing 747?
• Concord is 1350 mph / 610 mph 2.2 times faster
• Throughput of Concorde vs. Boeing 747 ?
• Boeing is 286,700 pmph / 178,200 pmph 1.6
  times faster

14
What affects performance?

• Execution time clock cycles x cycle time
• Clock cycles for a given program
• Instructions executed x Average CPI

15
Parameters to compare

• Table below shows execution time factors in
  different situations
• Differences determine relative performance

16
Tradeoffs

• Improving one parameter often worsens the other
• Micro-architects may have a choice between
• Reducing CPI and increasing clock cycle time, or
• Reducing clock cycle time and increasing CPI
• Compiler writers may have several choices that
  differ in instruction count and CPI.
• Example Consider the following High-level
  language code
• int a, b
• b a 5
• In generating machine code the compiler may have
  a choice between
• sll R2, R4, 2 1 cycle
• add R2, R2, R4 1 cycle
• or
• multiply R2, R4, 5 4 cycles