1. Rensselaer Polytechnic Institute CSCI-4210 – Operating Systems David Goldschmidt, Ph.D.

2.  A system is a part of the real world that we wish to analyze  Made up of autonomous entities interacting with one another  A model is an abstract representation of a system  We use models in analysis and design  Only relevant properties are included

3.  Simulate a system’s behavior using a model  Models have tunable input parameters  Observe the simulation; analyze output  Predict behavior of the real system by analyzing behavior of the model  Behavior of the model depends on time, input parameters, and events generated within the environment

4.  We can model numerous systems using the following components:  User: a single user of the system  Service: a node providing a service  Queue: an array of users waiting for service

5.  Mathematical models can be categorized as:  Deterministic: ▪ Behavior is predictable with 100% certainty  Stochastic: ▪ Behavior is uncertain, based on random events

6.  A stochastic model is one that incorporates uncertainty into its behavior  One or more attributes change their values according to a probability distribution

7.  A probability distribution specifies the probability of each value of some random variable  Uniform distribution:  All values are equally probable

8.  A probability distribution specifies the probability of each value of some random variable  Normal distribution:  Values are more probable at or near the mean  This will form a bell curve

9.  A probability distribution specifies the probability of each value of some random variable  Exponential distribution:  Values are times between events in a Poisson process in which events occur both continuously and independently, but at a constant average rate

10.  To study the performance of an operating system, we can:  Take measurements on the real system ▪ e.g. Unix time command  Run a simulation model  Apply an analytical model ▪ e.g. Queuing theory, exponential distributions, etc.

11.  External performance goals:  Minimize user response time  Maximize throughput ▪ Number of jobs completed per unit time  Minimize turnaround time ▪ Average time to complete jobs  Maximize fairness  Maximize degree of multiprogramming ▪ Number of processes supported without degradation

12.  Internal performance goals:  Maximize CPU utilization  Maximize disk utilization  Minimize disk access time  Enforce priorities  Minimize overhead ▪ e.g. time for scheduling algorithm, context switching  Avoid starvation of long-running jobs  Enforce real-time deadlines (sometimes)

13.  Some key performance measures:  Average number of jobs in the system  Average number of jobs waiting in a queue  Average time a job spends in the system  Average time a job spends in the queues  CPU utilization  Total number of jobs serviced (i.e. throughput)

14.  Processes are created by the operating system  Processes initially added to a job queue, which contains all processes waiting to enter the system  From the job queue, processes that are ready for execution are added to the ready queue

15.  A long-term scheduler (i.e. job scheduler) selects processes from the job queue, adding those processes to the ready queue  A short-term scheduler (i.e. CPU scheduler) selects processes from the ready queue and allocates time with the CPU

16.  The long-term scheduler is invoked infrequently

17.  The degree of multiprogramming of an operating system is defined as the number of processes in memory  In a stable operating system, the average process arrival rate equals the average process departure rate

18.  The short-term scheduler decides which process the CPU executes next  The dispatcher gives control of the CPU to the process selected by the CPU scheduler:  Performs context switch  Switches to user mode  Jumps to the proper location in the user program to resume program execution

19. the dispatcher operates here

20.  Processes alternate between CPU execution and I/O wait  A CPU burst is actual program execution that uses the CPU  An I/O burst is a blocked state  Each process starts and ends with a CPU burst

21.  Histogram of CPU burst time frequencies

22.  CPU scheduling requires an algorithm to determine which process to dispatch next  Scheduling algorithms include:  First-Come, First-Served (FCFS)  Shortest-Job-First (SJF)  Round-Robin (RR)  Priority  Multilevel Queue (MQ)

23.  Preemptive scheduling preempts a running process before its time slice expires  Or it preempts a process because its time slice has expired  Non-preemptive scheduling gives a process exclusive uninterrupted access to the CPU for the entirety of its execution process

24.  Compare scheduling algorithms by measuring  CPU utilization – keep CPU as busy as possible  Throughput – maximize the number of processes that complete their execution per unit time  Turnaround time – minimize the elapsed time to fully execute a particular process  Waiting time – minimize the elapsed time a process waits in the ready queue