1. Chapter 3 System Performance and Models

2. 2 3.1 Introduction A system is the part of the real world under study. Composed of a set of entities interacting among themselves and with the environment. A model is an abstract representation of a system. The system behavior is dependent on the input data and actions from the environment.

3. 3 Abstraction – The most important concept in analysis and design – A high-level description of a collection of objects – Only the relevant properties and features of the objects are included.

4. 4 System A system has: Structure Behavior The model of a system is simpler than the real system in its structure and behavior. But it should be equivalent to the system.

5. 5 Using Models A user can: Manipulate the model by supplying it with a set of inputs Observe its behavior or output Predict the behavior of the real system by analyzing the behavior of the model.

6. 6 Behavior of a Model Depends on: The passage of time Input data Events generated by the environment

7. 7 3.2 General Concepts of Model The most general categories of models are: Physical models (scale models) Graphical models Mathematical models. Mathematical models are the most flexible ones.

8. 8 3.3 Simple Models of Computer Systems Model of a Simple Batch System

9. 9 Simulation Results For every simulation run there are two types of output: Trace - sequence of events that occur during the simulation period Performance measures - summary statistics about the simulation.

10. 10 3.4 Performance of Computer Systems Measures that indicate how well (or bad) the system is being studied carrying out its functions, with respect to some aspects In studying a system, usually several performance measures are necessary.

11. 11 Approaches for Studying Performance Measurements on the real system Simulation models Analytical (mathematical) models

12. 12 Performance Study A complete performance study includes the definition of the following components: – A set of relevant objectives and goals – Performance metrics – System parameters – System factors – System workload parameter

13. 13 3.4.1 Performance Measures Average wait time: the average period that jobs wait in system since their arrival time Average throughput: average number of jobs completed in some specified time interval. CPU Utilization: the portion of time the CPU is used relative to total observation interval. Resource Utilization: the proportion of the interval the resource is used. Response time: the average time interval that the system takes to respond to a particular command or request of a user process.

14. 14 Average turnaround time: the average time interval that elapses from the time a job is submitted until the time the system writes the output results (also called sojourn time). Availability: the fraction of time that an external observer finds the system capable of carrying out some work. Reliability: the mean time between failures. Capacity: maximum achievable throughput under ideal workload conditions. Fairness: a measure of the variability of throughput across the various types of jobs or processes. Speedup: a factor of gains in speed usually achieved by adding more processors to a system. Performance Measures (cont’d)

15. 15 Reduce waiting periods for the processes Improve the processor utilization Maximize throughput Goal of Performance Study

16. 16 3.4.2 Workload and System Parameters (These are for the simulation model, Skip ) The performance measures depend on the current workload of the system The workload for a system can be characterized by another series of measures, which are made on the input to the system Errors in characterizing the workload may have serious consequences.

17. 17 Workload Parameters Inter-arrival time Task size I/O request rate I/O service rate Memory request

18. 18 System Parameters System memory Processor speed Number and type of processors Degree of multi-programming Length of time slice Number and type of I/O ports

19. Skip: 3.5 and 3.6 (Run simulator) 19

20. 20 3.7 System Capacity and Bottleneck The capacity of a system is determined by its maximum performance The nominal capacity of a system is given by the maximum achievable throughput under ideal conditions The usable capacity is the maximum throughput achievable under specified constraints.

21. 21 Bottleneck The computer system reaches capacity when one or more of its servers or resources reach a utilization close to 100%. The bottleneck of the system will be localized in the server or resource with a utilization close to 100%, while the other servers and resources each have utilization significantly below 100%.