Structure of Computer Systems

1
Structureof Computer Systems

• Dr. Zoltan Francisc Baruch
• Computer Science Department
• Technical University of Cluj-Napoca

2
Course Information

• Course grading
• 25 Laboratory ? colloquy
• 25 Project ? written report
• 50 Exam (15 mid-term, 35 final)
• Minimum grade for each activity 5
• Tests ? lecture
• Course website
• http//users.utcluj.ro/baruch/en/
• Teaching ? Structure of Computer Systems

3
Bibliography (1)

• Baruch, Z. F., Structure of Computer Systems,
  U.T.PRES, Cluj-Napoca, 2002, ISBN 973-8335-44-2

4
Bibliography (2)

• Baruch, Z. F., Structure of Computer Systems (in
  Romanian), Ed. Albastra, Cluj-Napoca, 2005, ISBN
  973-650-143-4

5
Bibliography (3)

• Baruch, Z. F., Structure of Computer Systems with
  Applications, U. T. PRES, Cluj-Napoca, 2003, ISBN
  973-8335-89-2

6
Contents of the Lecture

• 1. Introduction
• 2. Arithmetic-Logic Unit
• 3. Memory Systems
• 4. Pipeline Systems
• 5. RISC Architectures
• 6. Advanced Architectures

7
1. Introduction

• Performance Metrics
• Execution Time
• CPU Time
• MIPS
• MFLOPS
• Benchmark Programs
• Amdahls Law

8
Execution Time (1)

• Performance of a computer refers to
• Speed
• Hardware and software reliability
• The measure of performance execution time (tE)
• Response time (elapsed time) the time required
  to complete a task
• Includes memory accesses, input/output
  activities, and operating system overhead

9
Execution Time (2)

• CPU time the time the CPU is computing
• Does not include the time waiting for I/O
  operations
• Does not include the time for running other
  programs
• Can be divided into
• User CPU time
• System CPU time

10
Execution Time (3)

• Comparing the performance of two different
  computers, e.g., X and Y
• Computer X is faster than Y when the execution
  time of X is lower than that of Y for a given
  task
• Computer X is n faster than Y will mean

11
Execution Time (4)

• Since execution time tE is reciprocal of
  performance P
• The performance increase (n) will be
• Example 1.1

12
1. Introduction

• Performance Metrics
• Execution Time
• CPU Time
• MIPS
• MFLOPS
• Benchmark Programs
• Amdahls Law

13
CPU Time (1)

• CPU time (tCPU) can be expressed as
• CCPU number of CPU clock cycles to execute the
  program
• tC clock cycle time
• Another expression
• f clock frequency

14
CPU Time (2)

• The number of instructions executed may also be
  considered ? instruction count N
• The average number of clock cycles per
  instruction (CPI)
• CPU time can be defined as

15
CPU Time (3)

• or
• The number of total CPU cycles
• CPIi number of clock cycles for instruction i
• Ii number of times instruction i is executed

16
CPU Time (4)

• CPU time
• Overall no. of clock cycles per instruction
• Fi frequency of instruction i
• Example 1.2

17
1. Introduction

• Performance Metrics
• Execution Time
• CPU Time
• MIPS
• MFLOPS
• Benchmark Programs
• Amdahls Law

18
MIPS (1)

• The most important measure of performance the
  execution time of real programs
• A number of popular measures have been adopted
  for computer performance
• One of the performance metrics is called MIPS
  (Millions of Instructions Per Second)
• Indicates the number of average instructions
  that a computer can execute per second

19
MIPS (2)

• For a given program, MIPS is defined as
• N instruction count
• Assuming that tE tCPU ,
• We get

20
MIPS (3)

• Execution time expressed as a function of MIPS
• A metric defined similarly BIPS (Billions of
  Instructions Per Second) or GIPS
• The advantage of MIPS it is easy to understand,
  especially by a customer
• There are problems with using MIPS as a measure
  for comparison

21
MIPS (4)

• MIPS is dependent on the instruction set
• MIPS varies between programs on the same computer
  
• MIPS can vary inversely to performance
• Example of the last case a computer with
  optional floating-point coprocessor
• Programs using the coprocessor take less time,
  but have a lower MIPS rating
• Example 1.3

22
1. Introduction

• Performance Metrics
• Execution Time
• CPU Time
• MIPS
• MFLOPS
• Benchmark Programs
• Amdahls Law

23
MFLOPS (1)

• MIPS is not a good metric for computers that
  perform scientific and engineering computation
• It is important to measure the number of
  floating-point (FP) operations
• MFLOPS (Millions of Floating-point Operations Per
  Second), GFLOPS, TFLOPS, PFLOPS
• The formula

24
MFLOPS (2)

• NFP number of FP operations
• A MFLOPS rating is dependent on the machine and
  on the program
• Problems related to MFLOPS
• The set of FP operations is not consistent across
  machines
• The MFLOPS rating changes on
• The combination of integer and FP operations
• The combination of fast and slow FP operations

25
MFLOPS (3)

• The solution to both problems to use normalized
  FP operations
• Example for calculating the number of normalized
  FP operations according to the real operations in
  the source code

26
MFLOPS (4)

• The real FP operations give the native MFLOPS
  rating
• The normalized FP operations give the normalized
  MFLOPS rating
• The MIPS and MFLOPS metrics are useful for
  comparing members of the same architectural
  family
• These metrics are not appropriate for comparing
  computers with different instruction sets

27
MFLOPS (5)

• However, MFLOPS is used in some benchmark
  programs to evaluate the performance of
  supercomputers
• Example the Linpack benchmark
• Software library for numerical linear algebra
  (vector or matrix) operations
• HPL (High Performance Linpack) portable
  implementation of Linpack used for the TOP500 list

28
MFLOPS (6)

• TOP500 ranks the 500 most powerful publicly
  known computers in the world
• http//www.top500.org/
• The current list released in June 2008
• No. 1 in the list is the IBM Roadrunner
• Installed for the National Nuclear Security
  Administration (Department of Energy, USA) at Los
  Alamos National Laboratory
• Peak performance 1.026 PFLOPS

29
MFLOPS (7)

• The first hybrid supercomputer
• 6,562 dual-core AMD Opteron chips (1.8 GHz)
• 12,240 IBM PowerXCell 8i chips (3.2 GHz 12.8
  GFLOPS)
• Memory 98 TB (internal), 2 PB (external)
• Operating system Red Hat Linux
• Power consumption 3.9 MW ? one of the most
  energy efficient with 376 MFLOPS/W
• Occupies 278 racks

30
MFLOPS (8)

• The IBM Roadrunner installed at Los Alamos
  National Laboratory (New Mexico, USA)

31
Questions

• What are the differences between response time
  and CPU time?
• How can be expressed the CPU time based on the
  average number of clock cycles per instruction?
• What are the disadvantages of the MIPS metric?
• What are the problems related to the MFLOPS
  rating?