Survey
* Your assessment is very important for improving the workof artificial intelligence, which forms the content of this project
* Your assessment is very important for improving the workof artificial intelligence, which forms the content of this project
COMS 361 Computer Organization Title: Performance Date: 10/02/2004 Lecture Number: 3 1 Announcements • Homework 1 – Due next Tuesday 9/07/04 2 Review • Computer System Organization • Instruction Set Architecture (ISA) – Important from the programmers point of view – Allows families of processors • Hardware Set Architecture (HSA) – Defines operational components and their interconnections – An implementation of an ISA • Computer Organization – The implementation of an HSA 3 Outline • Trends • Performance – Measures – Comparisons 4 VLSI Trends: Moore’s Law • In 1965, Gordon Moore predicted that transistors would continue to shrink, allowing: – Doubled transistor density every 24 months – Doubled performance every 18 months • History has proven Moore right • But, is the end in sight? – Physical limitations – Economic limitations Gordon Moore Intel Co-Founder and Chairmain Emeritus 5 Microprocessor Trends (Log Scale) 6 Microprocessor Trends (Intel) Year Chip L transistors 1971 4004 10µm 2.3K 1974 8080 6µm 6.0K 1976 8088 3µm 29K 1982 80286 1.5µm 134K 1985 80386 1.5µm 275K 1989 80486 0.8µm 1.2M 1993 Pentium® 0.8µm 3.1M 1995 Pentium® Pro 0.6µm 15.5M 1999 Mobile PII 0.25µm 27.4 2000 Pentium® 4 0.18µm 42M 2002 Pentium® 4 (N) 0.13µm 55M 2003 Itanium® 2 (M) 0.13µm 410M 7 Microprocessor Trends 450 I2M Transistors (Millions) 400 350 300 Intel Motorola DEC/Compaq IBM 250 200 Alpha (R.I.P) 150 P4N, G5 100 50 0 1970 1975 1980 1985 1990 1995 2000 2005 8 Microprocessor Trends (Log Scale) 100 I2M Transistors (Millions) Alpha (R.I.P) 10 P4N, G5 G4 1 Intel Motorola DEC/Compaq 0.1 0.01 0.001 1960 1970 1980 1990 2000 9 DRAM Memory Trends (Log Scale) 1000 100 64 512 256 128 16 10 4 1 Size (Mb) 1 0.25 0.1 0.0625 0.01 1975 1980 1985 1990 1995 2000 2005 10 Performance Trends Vax 11/780 11 Summary - Technology Trends • Processor – Logic capacity – Clock frequency – Cost per function increases ~ 30% per year increases ~ 20% per year decreases ~20% per year • Memory – DRAM capacity: (4x every 3 years) – Speed: – Cost per bit: increases ~ 60% per year increases ~ 10% per year decreases ~25% per year • Disk – Storage capacity increases ~60% per year 12 Performance • Measure, Report, and Summarize – Understand major factors of performance – Software performance is a function of • Which instructions are used • Execution time of the instructions used 13 Performance • Performance measurement can be difficult – Complex software – Hardware performance enhancements – Different applications may perform differently on the same hardware • Graphics application • Computational • Database • Impossible to analytically compute a systems performance 14 Performance • An important attribute of a system – Purchasers – Designers – Sales people • Stretch the truth • Performance claims may be meaningless for your application • Focus – Why software performs as it does – How does the instruction set affect performance – What hardware features are responsible for improved performance 15 Performance: A Matter of Perspective • What does this mean? – One computer performs better than another? • Subtle question • Difficult to answer without an accurate meaning of the term performance • Example of a performance issue 16 Performance: A Matter of Perspective Airplane Passengers Boeing 737-100 Boeing 747 BAC/Sud Concorde Douglas DC-8-50 101 470 132 146 Range (mi) 630 4150 4000 8720 Speed (mph) 598 610 1350 544 • Which plane has the best performance? – Concord: fastest – DC-8: longest range – 747: largest capacity 17 Performance: A Matter of Perspective • Consider speed as a performance measure – Speed is still not precise enough – Concord: fastest plane for a individual – 747: fastest plane for transporting 450 people • Same program running on two computers – First machine done is the fastest, right? – Shared system: • The most tasks completed in the least amount of time has the better performance – Response time: individual user – Throughput: system manager 18 Performance Equation • Selfish motivation: – Focus on response time • Often faster response time means more throughput – Maximize performance by minimizing response time 1 Performanc e x Execution time x – Performance of machine X is better than the performance of machine Y if: Performanc e x Performanc e y 1 1 Execution time x Execution time y Execution time y Execution time x 19 Performance Equation • Performancex > Performancey • Means the Execution time(x) < Execution (y) • How much better one machine wrt another – relative performance measure of two machines Performanc e x n Performanc e y – If X is n times faster than Y, the execution time on Y is n times longer than on X Execution time y Exectution time x n 20 Measuring Performance • Time to execute a program is our measure • Which time? – Obvious measures: • Wall-clock time, response time, or elapsed time • The total time for task completion • Time-sharing (common) – Elapsed time – Time to complete our task • CPU time – The time the CPU spends on our task • User time (time executing the instructions in our program) • System time (OS things: opening files, networking, …) 21 Measuring Performance • Use the time command to determine – user time – system time – elapsed time • Combine user and system time for a macroscopic measure of performance • Designers – How fast can the hardware perform basic functions • Moving data, adding, multiplying – Most computers have a constant speed clock that determines when events happen • Clock – Ticks, cycles, period, rate, … – Heartbeat of the processor 22 Measuring Performance t 1 clock cycle or clock period (measured in time) 0.25 ns, 250 ps clock rate 1 clock period 4 GHz • Clock rate – Measured as the number of clock cycles per unit time 23 Orders of Magnitude thousandths 10-3 milli, m millionths 10-6 micro, u billionths 10-9 nano, n trillionths 10-12 pico, p • Disk access: ms • Memory access: us • Clock cycle time: ns, ps 24 Orders of Magnitude • • • • thousands 103 kilo, K millions 106 mega, M billions 109 giga, G trillions 1012 tera, T Size of L1 cache: k Size of Main memory: M, G Size of a disk drive: G Size of a disk farm: T 25 Orders of Magnitude Scientific Units Computer Science Units 103 = 1,000 thousands kilo, K 210 = 1024 106 = 1,000,000 millions mega, M 220 = 1,048,576 109 = 1,000,000,000 billions giga, G 230 = 1,073,741,824 1012 = 1,000,000,000,000 trillions tera, T 240 = 1,099,511,627,776 26 Clock/execution time • Relation between clock rate and execution time – Tells what happens to execution time if the rate changes CPU time number of clock cycles * clock period program CPU time number of clock cycles clock rate • Increase performance – Reduce the number of clocks to execute the program – Reduce the clock period • Increase the clock rate 27 Example • Our favorite program runs in 10 seconds on computer A, which has a 400 Mhz clock. We are trying to help a computer designer build a new machine B, that will run this program in 6 seconds. The designer can use new (or perhaps more expensive) technology to substantially increase the clock rate, but has informed us that this increase will affect the rest of the CPU design, causing machine B to require 1.2 times as many clock cycles as machine A for the same program. What clock rate should we tell the designer to target? • Don't Panic, can easily work this out from basic principles 28 Example • A given program will require – some number of instructions (machine instructions) – some number of cycles – some number of seconds • We have a vocabulary that relates these quantities: – cycle time (seconds per cycle) – clock rate (cycles per second) – CPI (cycles per instruction) a floating point intensive application might have a higher CPI – MIPS (millions of instructions per second) this would be higher for a program using simple instructions 29 Performance • Performance is determined by execution time • Do any of the other variables equal performance? – – – – – # of cycles to execute program? # of instructions in program? # of cycles per second? average # of cycles per instruction? average # of instructions per second? • Common pitfall: thinking one of the variables is indicative of performance when it really isn’t 30 CPI Example • Suppose we have two implementations of the same instruction set Architecture (ISA). • For some program, • Machine A has a clock cycle time of 10 ns. and a CPI of 2.0 • Machine B has a clock cycle time of 20 ns. and a CPI of 1.2 31 CPI Example • What machine is faster for this program, and by how much? • If two machines have the same ISA which of our quantities (e.g., clock rate, CPI, execution time, # of instructions, MIPS) will always be identical? 32 # of Instructions Example • A compiler designer is trying to decide between two code sequences for a particular machine. Based on the hardware implementation, there are three different classes of instructions: Class A, Class B, and Class C, and they require one, two, and three cycles (respectively) • The first code sequence has 5 instructions: – 2 of A, 1 of B, and 2 of C • The second sequence has 6 instructions: – 4 of A, 1 of B, and 1 of C 33 # of Instructions Example • Which sequence will be faster? • How much? • What is the CPI for each sequence? 34 MIPS example • Two different compilers are being tested for a 100 MHz. machine with three different classes of instructions: Class A, Class B, and Class C, which require one, two, and three cycles (respectively). Both compilers are used to produce code for a large piece of software. The first compiler's code uses – 5 million Class A instructions – 1 million Class B instructions – 1 million Class C instructions 35 MIPS example • The second compiler's code uses – 10 million Class A instructions – 1 million Class B instructions – 1 million Class C instructions • Which sequence will be faster according to MIPS? • Which sequence will be faster according to execution time? 36 Benchmarks • Performance best determined by running a real application – Use programs typical of expected workload – Or, typical of expected class of applications e.g., compilers/editors, scientific applications, graphics, etc. • Small benchmarks – nice for architects and designers – easy to standardize – can be abused 37 Benchmarks • SPEC (System Performance Evaluation Cooperative) – companies have agreed on a set of real program and inputs – can still be abused (Intel’s “other” bug) – valuable indicator of performance (and compiler technology) 38 SPEC ‘89 • Compiler “enhancements” and performance 800 700 SPEC performance ratio 600 500 400 300 200 100 0 gcc espresso spice doduc nasa7 li eqntott matrix300 fpppp tomcatv Benchmark Compiler Enhanced compiler 39 SPEC ‘95 Benchmark go m88ksim gcc compress li ijpeg perl vortex tomcatv swim su2cor hydro2d mgrid applu trub3d apsi fpppp wave5 Description Artificial intelligence; plays the game of Go Motorola 88k chip simulator; runs test program The Gnu C compiler generating SPARC code Compresses and decompresses file in memory Lisp interpreter Graphic compression and decompression Manipulates strings and prime numbers in the special-purpose programming language Perl A database program A mesh generation program Shallow water model with 513 x 513 grid quantum physics; Monte Carlo simulation Astrophysics; Hydrodynamic Naiver Stokes equations Multigrid solver in 3-D potential field Parabolic/elliptic partial differential equations Simulates isotropic, homogeneous turbulence in a cube Solves problems regarding temperature, wind velocity, and distribution of pollutant Quantum chemistry Plasma physics; electromagnetic particle simulation 40 SPEC ‘95 10 10 9 9 8 8 7 7 6 6 SPECfp SPECint Does doubling the clock rate double the performance? Can a machine with a slower clock rate have better performance? 5 5 4 4 3 3 2 2 1 1 0 0 50 100 150 Clock rate (MHz) 200 250 Pentium Pentium Pro 50 100 150 Clock rate (MHz) 200 250 Pentium Pentium Pro 41 Amdahl's Law Execution Time After Improvement = Execution Time Unaffected +( Execution Time Affected / Amount of Improvement ) • Example: "Suppose a program runs in 100 seconds on a machine, with multiply responsible for 80 seconds of this time. How much do we have to improve the speed of multiplication if we want the program to run 4 times faster?" 42 Example • Suppose we enhance a machine making all floating-point instructions run five times faster. If the execution time of some benchmark before the floating-point enhancement is 10 seconds, what will the speedup be if half of the 10 seconds is spent executing floating-point instructions? • We are looking for a benchmark to show off the new floating-point unit described above, and want the overall benchmark to show a speedup of 3. One benchmark we are considering runs for 100 seconds with the old floatingpoint hardware. How much of the execution time would 43 Remember • Performance is specific to a particular program/s – Total execution time is a consistent summary of performance • For a given architecture performance increases come from: – increases in clock rate (without adverse CPI affects) – improvements in processor organization that lower CPI – compiler enhancements that lower CPI and/or instruction count 44