CS 501: Software EngineeringAdministrationPerformance of Computer SystemsMoore's LawMoore's Law: Rules of ThumbMoore's Law and System DesignParkinson's LawFalse Assumptions from the PastMoore's Law and the Long TermSlide 10Predicting System PerformanceUnderstand the Interactions between Hardware and SoftwareSlide 13Understand Interactions between Hardware and SoftwareLook for BottlenecksLook for Bottlenecks: UtilizationTechniques for Eliminating BottlenecksMathematical Models: QueuesQueuesMathematical ModelsBehavior of Queues: UtilizationSimulationTimescaleMeasurements on Operational SystemsExample: Performance of Disk Array1CS 501 Spring 2003CS 501: Software EngineeringLecture 23Performance of Computer Systems2CS 501 Spring 2003AdministrationFinal presentationsSign up now. Available time slots are on the Web site.3CS 501 Spring 2003Performance of Computer SystemsIn most computer systemsThe cost of people is much greater than the cost of hardwareYet performance is importantFuture loads may be much greater than predictedA single bottleneck can slow down an entire system4CS 501 Spring 2003Moore's LawOriginal version: The density of transistors in an integrated circuit will double every year. (Gordon Moore, Intel, 1965) Current version:Cost/performance of silicon chips doubles every 18 months.5CS 501 Spring 2003Moore's Law: Rules of ThumbPlanning assumptions:Every year: cost/performance of silicon chips improves 25% cost/performance of magnetic media improves 30%10 years = 100:120 years = 10,000:16CS 501 Spring 2003Moore's Law and System DesignDesign system: 2003Production use: 2006Withdrawn from production: 2016Processor speeds: 1 1.9 28Memory sizes: 1 1.9 28Disk capacity: 1 2.2 51System cost: 1 0.4 0.017CS 501 Spring 2003Parkinson's LawOriginal: Work expands to fill the time available. (C. Northcote Parkinson)Planning assumptions:(a) Demand will expand to use all the hardware available.(b) Low prices will create new demands.(c) Your software will be used on equipment that you have not envisioned.8CS 501 Spring 2003False Assumptions from the PastUnix file system will never exceed 2 Gbytes (232 bytes).AppleTalk networks will never have more than 256 hosts (28 bits).GPS software will not last 1024 weeks.Nobody at Dartmouth will ever earn more than $10,000 per month.etc., etc., .....9CS 501 Spring 2003Moore's Law and the Long Term1965What level?200310CS 501 Spring 2003Moore's Law and the Long Term1965When?What level?2003?Within your working life?11CS 501 Spring 2003Predicting System Performance• Mathematical models• Simulation• Direct measurement• Rules of thumbAll require detailed understanding of the interaction between software and systems.12CS 501 Spring 2003Understand the Interactions between Hardware and SoftwareExample: execution of http://www.cs.cornell.edu/Client Serversdomain name serviceTCP connectionHTTP get13CS 501 Spring 2003Understand the Interactions between Hardware and Software:Thread :Toolkit :ComponentPeer target:HelloWorldrunrun callbackLoophandleExposepaint14CS 501 Spring 2003DecompressStream audioStream videoforkjoinstart statestop stateUnderstand Interactions between Hardware and Software15CS 501 Spring 2003Look for BottlenecksPossible areas of congestionNetwork loadDatabase accesshow many joins to build a record?Locks and sequential processingCPU performance is rarely a factor, except in mathematical algorithms. More likely bottlenecks are:Reading data from diskMoving data from memory to CPU.16CS 501 Spring 2003Look for Bottlenecks: Utilizationutilization = mean service timemean inter-arrival timeWhen the utilization of any hardware component exceeds 30%, be prepared for congestion.17CS 501 Spring 2003Techniques for Eliminating BottlenecksSerial and Parallel ProcessingSingle thread v. multi-threade.g., Unix forkGranularity of locks on datae.g., record lockingNetwork congestione.g., back-off algorithms18CS 501 Spring 2003Mathematical Models: Queuesarrive wait in line service departSingle server queue19CS 501 Spring 2003Queuesarrive wait in lineservicedepartMulti-server queue20CS 501 Spring 2003Mathematical ModelsQueueing theoryGood estimates of congestion can be made for single-server queues with: • arrivals that are independent, random events (Poisson process)• service times that follow families of distributions (e.g., negative exponential, gamma)Many of the results can be extended to multi-server queues.21CS 501 Spring 2003Behavior of Queues: Utilizationmeandelayutilization1022CS 501 Spring 2003SimulationModel the system as set of states and eventsadvance simulated time determine which events occurred update state and event listrepeatDiscrete time simulation: Time is advanced in fixed steps (e.g., 1 millisecond)Next event simulation: Time is advanced to next eventEvents can be simulated by random variables (e.g., arrival of next customer, completion of disk latency)23CS 501 Spring 2003TimescaleOperations per secondCPU instruction: 1,000,000,000Disk latency: 60 read: 25,000,000 bytesNetwork LAN: 10,000,000 bytesdial-up modem: 6,000 bytes24CS 501 Spring 2003Measurements on Operational Systems• Benchmarks: Run system on standard problem sets, sample inputs, or a simulated load on the system.• Instrumentation: Clock specific events.If you have any doubt about the performance of part of a system, experiment with a simulated load.25CS 501 Spring 2003Example: Performance of Disk ArrayEach transaction must:wait for specific disk platterwait for I/O channelsignal to move heads on disk platterwait for I/O channelpause for disk rotationread dataClose agreement between: results from queuing theory, simulation, and direct measurement (within
View Full Document
Unlocking...