Description
Project specifications
In this project, you will implement a rudimentary simulation of an operating system. The initial focus will be on processes, assumed to be resident in memory, waiting to use the CPU. Memory and the I/O subsystem will not be covered in depth in this project.
Conceptual Design
A process is defined as a program in execution. For this assignment, processes are in one of the following three states:
READY: in the ready queue, ready to use the CPU
RUNNING: actively using the CPU
BLOCKED: blocked on I/O
Processes in the READY state reside in a queue called the ready queue. This queue is ordered based on a configurable CPU scheduling algorithm. In this first assignment, there are four algorithms to implement: shortest job first (SJF); shortest remaining time (SRT); first come first served (FCFS); and round robin (RR). Note that if you use a large enough time slice, RR essentially becomes the FCFS algorithm.
All four of these algorithms will be simulated for the same set of processes, which will allow for a comparative analysis. As such, when you run your program, all four algorithms are simulated.
And in general, when a process reaches the front of the queue and the CPU is free to accept the next process, the given process enters the RUNNING state and starts executing its CPU burst.
After the CPU burst is completed, if the process does not terminate, the process enters the BLOCKED state, waiting for an I/O operation to complete (e.g., waiting for data to be read in from a file). When the I/O operation completes, depending on the scheduling algorithm, the process either (1) returns to the READY state and is added to the ready queue or (2) preempts the currently running process and switches into the RUNNING state.
Note that preemptions occur only for some algorithms, i.e., for SRT and RR. Each algorithm is summarized on the next page.
Shortest Job First (SJF)
In SJF, processes are stored in the ready queue in order of priority based on their CPU burst times. More specifically, the process with the shortest CPU burst time will be selected as the next process executed by the CPU.
Shortest Remaining Time (SRT)
The SRT algorithm is a preemptive version of the SJF algorithm. In SRT, when a process arrives, before it enters the ready queue, if it has a CPU burst time that is less than the remaining time of the currently running process, a preemption occurs. When such a preemption occurs, the currently running process is added back to the ready queue.
First Come, First Served (FCFS)
The FCFS algorithm is a non-preemptive algorithm in which processes line up in the ready queue, waiting to use the CPU. This is your baseline algorithm (and may be implemented as RR with an infinite time slice).
Round Robin (RR)
The RR algorithm is essentially the FCFS algorithm with predefined time slice tslice. Each process is given tslice amount of time to complete its CPU burst. If this time slice expires, the process is preempted and added to the end of the ready queue (though see the rradd parameter described below).
If a process completes its CPU burst before a time slice expiration, the next process on the ready queue is immediately context-switched into the CPU.
Skipping preemptions in RR
For your simulation, if a preemption occurs but there are no other processes on the ready queue, do not perform a context switch. For example, if process G is using the CPU and the ready queue is empty, if process G is preempted by a time slice expiration, do not context-switch process G back to the empty queue. Instead, keep process G running with the CPU and do not count this as a context switch. In other words, when the time slice expires, check the queue to determine if a context switch should occur.
Simulation configuration
The key to designing a useful simulation is to provide a number of configurable parameters to the user. This allows you to simulate and tune for a variety of scenarios, e.g., a large number of CPU-bound processes, a variety of average process interarrival times, multiple CPUs, etc.
Therefore, define the following simulation parameters as tunable constants within your code, all of which will be given as command-line arguments:
argv[1]: We will use a random number generator to determine the interarrival times of CPU bursts. Since we can only generate pseudo-random numbers, the first command-line argument, s, serves as the seed for the random number generator. To ensure predictability and repeatability, use srand48() with this given seed before each scheduling algorithm and drand48() to obtain the next value in the range [0.0,1.0). For other languages, implement an equivalent 48-bit linear congruential generator, as described in the man page for these functions.
argv[2]: To determine interarrival times, we will use an exponential distribution; therefore, the second command-line argument is parameter λ. Remember that λ1 will be the average random value generated (e.g., if λ = 0.01, then the average should be appoximately 100). See the exp-random.c example. Use the formula in the code, i.e., -log( r ) / lambda.
argv[3]: As part of the exponential distribution, the third command-line argument represents the upper bound for valid pseudo-random numbers. Remember that this threshold is used to avoid values far down the long tail of the exponential distribution. As an example, if this is set to 3000, all generated values above 3000 should be skipped. For cases in which this value is used in the ceiling function (see the next page), be sure the ceiling is still valid according to this upper bound.
argv[4]: Define n as the number of processes to simulate. Process IDs are assigned in alphabetical order A through Z. Therefore, at most you will have 26 processes to simulate.
argv[5]: Define tcs as the time, in milliseconds, that it takes to perform a context switch. Remember that a context switch occurs each time a process leaves the CPU and is replaced by another process. Note that the first half of the context switch time (i.e., ) is the time required to remove the given process from the CPU; the second half of the context switch time is the time required to bring the next process in to use the CPU. Therefore, expect tcs to be a positive even integer.
argv[6]: For the SJF and SRT algorithms, since we cannot know the actual CPU burst times beforehand, we will rely on estimates determined via exponential averaging (as discussed in (v1.2) Lecture 9). As such, this command-line argument is the constant α. And note that the initial guess for each process is τ0 = λ1. When calculating τ values, use the “ceiling” function for all calculations.
argv[7]: For the RR algorithm, define the time slice value, tslice, measured in milliseconds
argv[8]: Also for the RR algorithm, define whether processes are added to the end or the beginning of the ready queue when they arrive or complete I/O. This optional command-line argument, rradd, is set to either BEGINNING or END, with END being the default behavior.
Pseudo-random numbers and predictability
A key aspect of this assignment is to compare the results of each of the simulated algorithms with one another given the same conditions. To ensure each CPU scheduling algorithm is given the same set of processes, you will need to carefully follow the algorithm below to identify the set of processes. This algorithm should be fully executed before applying any of the scheduling algorithms. For each of the n processes, in order A through Z:
- Identify the initial process arrival time as the next random number in the sequence; more specifically, use the “floor” of the next random number generated as shown in the posted exp-random.c example (note that you could have a 0 arrival time)
- Identify the number of CPU bursts for the given process as the next random number multiplied by 100 and truncated (e.g., 0.454928 becomes 45, 0.087188 becomes 8, etc.), then incremented by 1; this should obtain a random integer in the range [1,100]
- For each of these CPU bursts, identify the actual CPU burst time and the I/O burst time as the next two random numbers in the sequence, obtained by using the ceiling (i.e., ceil()) of the next random number generated as shown in the exp-random.c example; for the last CPU burst, do not generate an I/O burst time (since each process ends with a final CPU burst)
After you simulate each scheduling algorithm, you must reset the simulation back to the initial set of processes and set your elapsed time back to zero. More specifically, you must re-seed your random number generator to ensure the same set of processes and interarrival times.
Note that there may be times during your simulation in which the simulated CPU is idle because all processes are busy performing I/O. Also, when all processes terminate, your simulation ends.
Handling “ties”
For events that occur at the same time, use the following order to break the “tie”: (a) CPU burst completion; (b) I/O burst completion (i.e., back to the ready queue); and then (c) new process arrival.
All “ties” that occur within one of these three categories are to be broken using process ID order. As an example, if processes Q and T happen to both finish with their I/O at the same time, process Q wins this “tie” (because Q is alphabetically before T) and is added to the ready queue before process T.
Be sure you do not implement any additional logic for the I/O subsystem. In other words, there are no I/O queues to implement in this project.
CPU burst time
CPU burst times are randomly generated for each process that you simulate (see algorithm above). CPU burst time is defined as the amount of time a process is actually using the CPU. Therefore, this measure does not include context switch times.
Turnaround time
Turnaround times are to be measured for each process that you simulate. Turnaround time is defined as the end-to-end time a process spends in executing a single CPU burst.
More specifically, this is measured from process arrival time through to when the CPU burst is completed and the process is switched out of the CPU. Therefore, this measure includes the second half of the initial context switch in and the first half of the final context switch out, as well as any other context switches that occur while the CPU burst is being completed (i.e., due to preemptions).
Wait time
Wait times are to be measured for each process that you simulate. Wait time is defined as the amount of time a process spends waiting to use the CPU, which equates to the amount of time the given process is actually in the ready queue. Therefore, this measure does not include context switch times that the given process experiences (i.e., only measure the time the given process is actually in the ready queue).
More specifically, a process leaves the ready queue when it is switched into the CPU, which takes half of context switch time tcs. Likewise, a preempted process leaves the CPU and enters the ready queue after the first half of tcs.
Required terminal output
Your simulator should keep track of elapsed time t (measured in milliseconds), which is initially zero for each scheduling algorithm. As your simulation proceeds, t advances to each “interesting” event that occurs, displaying a specific line of output that describes each event.
Your simulator must display results for each of the four algorithms you simulate. For each algorithm, display a summary of the “pseudo-randomly” generated processes (which should be the same for each algorithm), then the “interesting” events from time 0 through time 999, followed only by process termination events and the final end-of-simulation event. See example output files posted on Submitty.
Your simulator must display a line of output for each “interesting” event that occurs using the format shown below. Note that the contents of the ready queue are shown for each event.
time <t>ms: <event-details> [Q <queue-contents>] And the “interesting” events are:
Start of simulation
Process arrival
Process starts using the CPU
Process finishes using the CPU (i.e., completes a CPU burst)
Process has its τ value recalculated (i.e., after a CPU burst completion)
Process preemption
Process starts performing I/O
Process finishes performing I/O
Process terminates by finishing its last CPU burst
End of simulation
The “process arrival” event occurs every time a process arrives, i.e., based on the initial arrival time and when a process completes I/O. In other words, processes “arrive” within the subsystem that consists of the CPU and the ready queue.
The “process preemption” event occurs every time a process is preempted by a time slice expiration (in RR) or by an arriving process (in SRT). When a preemption occurs, a context switch occurs (unless for RR there are no available processes in the ready queue).
Note that when your simulation ends, you must display that event as shown below.
time <t>ms: Simulator ended for <algorithm> [Q empty]
Be sure that you still include the process removal time (i.e., half the context switch time) for this last process.
Required output file
In addition to the above output (which should be sent to stdout), generate an output file called simout.txt that contains statistics for each simulated algorithm. The file format is shown below (with # as a placeholder for actual numerical data). Round to exactly three digits after the decimal point for your averages.
Algorithm SJF
— average CPU burst time: #.### ms
— average wait time: #.### ms
— average turnaround time: #.### ms
— total number of context switches: #
— total number of preemptions: #
Algorithm SRT
— average CPU burst time: #.### ms
— average wait time: #.### ms
— average turnaround time: #.### ms
— total number of context switches: #
— total number of preemptions: #
Algorithm FCFS
— average CPU burst time: #.### ms
— average wait time: #.### ms
— average turnaround time: #.### ms
— total number of context switches: #
— total number of preemptions: #
Algorithm RR
— average CPU burst time: #.### ms
— average wait time: #.### ms
— average turnaround time: #.### ms
— total number of context switches: #
— total number of preemptions: #
Note that averages are averaged over all executed CPU bursts. Also note that to count the number of context switches, you should count the number of times a process starts using the CPU.
Error handling
If improper command-line arguments are given, report an error message to stderr and abort further program execution. In general, if an error is encountered, display a meaningful error message on stderr, then abort further program execution.
Error messages must be one line only and use the following format:
ERROR: <error-text-here>
