Measuring Performance by Irfanullah
- 1. Measuring Performance By Irfanullah MS (Computer Science) University Of Peshawar
- 2. In This Demonstration What is Performance? Performance Measurement Methods MIPS CPI / IPC Choosing Programs to Evaluate Performance Benchmark Suits Speedup Quantitative Principles of Computer Design Future Work
- 3. Performance Performance means How quickly a given system executes a program/ a set of programs Two aspects Response Time The time between the start and the completion of an event– execution time. Throughput The total amount of work done in a given time
- 4. MIPS Millions of Instructions Per Second An early measure of computer performance. MIPS = No. of instruction executed in a program Time required to run the program typically expressed in millions of instructions per second. Out Of Use It does not account for the fact that different systems often require different no.s of instructions to implement a given program. A computer’s MIPS ratings does not tell you about how many instructions it require to perform a given task.
- 5. CPI / IPC CPI Cycle Per Instruction No of clock cycles required to execute each instruction CPI = No. of clock cycles required No. of instructions executed IPC Instructions Per Cycle No. of instructions executed per cycle IPC = No of Instructions executed No of clock cycles required Large IPC tends to indicate good performance & large CPI indicates poor performance
- 6. Choosing Programs to Evaluate Performance Five levels of programs used to evaluate performance are Real applications Compilers, Text-processing S/W like Word, and Photoshop. Modified ( or scripted) applications To enhance portability or to focus on one particular aspect of system performance Kernels To isolate performance of individual features of a machine. Toy Benchmarks Typically between 10- 100 lines if code and produce a result the user already knows. Synthetic benchmarks Try to match the average frequency of operations and operands of a large set of programs.
- 7. Benchmark Suits A set of programs that are believed to be typical of the programs that will be run on the system System is checked for how long it take to execute all of the programs in the suite. For example SPEC benchmark by Standard Performance Evaluation Cooperation Other benchmarks are: Business Winstone CC Windows Winbench
- 8. Speedup Speedup tells how the performance of an architecture changes as different improvements are made to architecture Ratio of the execution times before and after a change is made Speedup = Execution time old(before) Execution time new(after )
- 9. Quantitative Principles of Computer Design (Continued) Some principles that are useful in design and analysis of computers Make The Common Case Faster Amdahl’s Law CPU Performance Equation Principles Of Locality Taking Advantage of Parallelism
- 10. Make the Common Case Fast In making a design trade-off, Favor the frequent case over the infrequent case because The frequent case is often simpler and can be done faster than the infrequent case e.g. when multiplying two numbers, overflow is the infrequent case, no overflow is the frequent. However, system will slow down when overflow occurs. We have to decide what the frequent case is and how much performance can be improved by making that case faster Amdahl’s law quantify this principle
- 11. Amdahl’s Law Impact of a given performance improvement on overall performance depends on How much the improvement improves the performance ( when in use) How often the improvement is in use Execution Time new = Execution Time old Fraction unused +Fraction used Speedup used Where Fraction unused = Fraction of time( not instructions) that the improvement is not in use. Fraction used = Fraction of time that the improvement is in use Speedup used = Speedup that occurs when the improvement is used
- 12. Amdahl’s Law Continued) Fraction used and Fraction unused are computed using the execution time before the modification is applied Speedup can be defined as Speedup = Execution time old(before) Execution time new(after) 1 = Fraction unused +Fraction used Speedup used
- 13. Amdahl’s Law (Continued) The law states “ Performance improvement to be gained from using some faster mode of execution is limited by the fraction of the time the faster mode can be used” Speedup = Performance for entire task using the enhancement Performance for entire task without using the enhancement Or Speedup = Execution time for the entire work without using the enhancement Execution time for entire task using the enhancement when possible Speedup = Execution time old(before) Execution time new(after) = 1 (1- Fraction enhanced )+Fraction enhanced Speedup enhanced
- 14. An Example Problem. An enhancement is made to a processor of a server system. The new CPU time is 20 times faster than original. If the original CPU is busy with computation 35% of the time and is waiting for I/O 65% of the time, what is the overall speedup gained by incorporating the enhancement. Solution. Fraction enhanced = 35/100=.35. Speedup enhanced = 20. = 1 (1- 0.35 )+0.35 20 = 1.498173
- 15. CPU Performance Equation CPU time = CPU clock cycles for a program x Clock cycle time -------(1) Clock Cycle Time = 1 Clock rate (1) Implies CPU time = CPU clock cycles for a program -------------------(2) Clock rate Now, CPI = CPU clock cycles for a program Instruction Count CPU clock cycle for a program = CPI x IC Now (1) implies CPU time = CPI x IC x Clock Cycle time ------------(3)
- 16. CPU Performance Equation (Continued) By putting values equation (3) becomes CPU time = Instructions x clock cycles x Seconds Program x Instructions x Clock cycles = Seconds Program Unfortunately, it is difficult to change one parameter in complete isolation from others because the basic technologies involved in changing each characteristic are interdependent: Clock Cycle Time = H/W technology and organization CPI = Organization and instruction set architecture Instruction Count = Instruction set architecture and compiler technology
- 17. Principles of Locality Programs tend to reuse data and instructions they have used recently A program spends 90% of its execution time in only 10% of the code We can predict with reasonable accuracy what instructions and data a program will use in the near future based on its accesses in the recent past. Two types of locality Temporal locality States that recently accessed items are likely to be accessed in the near future. Spatial Locality Items whose addresses are near one another tend to be referenced close together in time.
- 18. Taking Advantage of Parallelism One of the most important methods for improving performance – It is obtained through Pipelining – to overlap the execution of instructions, so as to reduce the total time to complete a sequence of instructions.
- 19. Future Work Research is carried out on various aspects in order to improve performance and reduce cost
- 20. Thanks Queries And Suggestions