SRTF scheduling program in C++ with explanation
Similar Posts You May Be Interested In:
- Reader Writer Problem Code in C
- FCFS Scheduling Algorithm Code
- Nonpreemptive Priority Scheduling Code
- Preemptive Priority Scheduling Code
- SJF Scheduling Code
- SRTF Scheduling Code
- Round Robin Scheduling Code
What is SRTF Scheduling Algorithm
- SRTF stands for Shortest Remaining Time First.
- The process that has least burst time gets the CPU first.
- The processes gets serviced by the CPU in order of their burst time in ascending order.
- SRTF is preemptive. A process that is running on the CPU can be removed if a new process arrives in ready queue with lower burst time than current process.
- SRTF is basically preemtive version of SJF.
If you want to understand more about SRTF algorithm with example, watch the below video.
Implementing SRTF Algorithm in C++
Input
- Number of processes
- Arrival time of each process. If all process arrives at the same time, this can be set to 0 for all processes.
- Burst time of each process
Output
- Start Time(ST), Completion Time(CT), Turnaround Time(TAT), Waiting Time(WT) and Response Time(RT) for each process.
- Average turnaround Time, average waiting time and average response time.
- Throughput and cpu utilization.
Algorithm
completed = 0
current_time = 0
while(completed != n) {
find process with minimum burst time among process that are in ready queue at current_time
if(process found) {
if(process is getting CPU for the first time) {
start_time = current_time
}
burst_time = burst_time - 1
current_time = current_time + 1
if(burst_time == 0) {
completion_time = current_time
turnaround_time = completion_time - arrival_time
waiting_time = turnaround_time - burst_time
response_time = start_time - arrival_time
mark process as completed
completed++
}
}
else {
current_time++
}
}
If you want to understand how these formulas are derived, watch the below video.
#include <iostream>
#include <algorithm>
#include <iomanip>
#include <string.h>
using namespace std;
struct process {
int pid;
int arrival_time;
int burst_time;
int start_time;
int completion_time;
int turnaround_time;
int waiting_time;
int response_time;
};
int main() {
int n;
struct process p[100];
float avg_turnaround_time;
float avg_waiting_time;
float avg_response_time;
float cpu_utilisation;
int total_turnaround_time = 0;
int total_waiting_time = 0;
int total_response_time = 0;
int total_idle_time = 0;
float throughput;
int burst_remaining[100];
int is_completed[100];
memset(is_completed,0,sizeof(is_completed));
cout << setprecision(2) << fixed;
cout<<"Enter the number of processes: ";
cin>>n;
for(int i = 0; i < n; i++) {
cout<<"Enter arrival time of process "<<i+1<<": ";
cin>>p[i].arrival_time;
cout<<"Enter burst time of process "<<i+1<<": ";
cin>>p[i].burst_time;
p[i].pid = i+1;
burst_remaining[i] = p[i].burst_time;
cout<<endl;
}
int current_time = 0;
int completed = 0;
int prev = 0;
while(completed != n) {
int idx = -1;
int mn = 10000000;
for(int i = 0; i < n; i++) {
if(p[i].arrival_time <= current_time && is_completed[i] == 0) {
if(burst_remaining[i] < mn) {
mn = burst_remaining[i];
idx = i;
}
if(burst_remaining[i] == mn) {
if(p[i].arrival_time < p[idx].arrival_time) {
mn = burst_remaining[i];
idx = i;
}
}
}
}
if(idx != -1) {
if(burst_remaining[idx] == p[idx].burst_time) {
p[idx].start_time = current_time;
total_idle_time += p[idx].start_time - prev;
}
burst_remaining[idx] -= 1;
current_time++;
prev = current_time;
if(burst_remaining[idx] == 0) {
p[idx].completion_time = current_time;
p[idx].turnaround_time = p[idx].completion_time - p[idx].arrival_time;
p[idx].waiting_time = p[idx].turnaround_time - p[idx].burst_time;
p[idx].response_time = p[idx].start_time - p[idx].arrival_time;
total_turnaround_time += p[idx].turnaround_time;
total_waiting_time += p[idx].waiting_time;
total_response_time += p[idx].response_time;
is_completed[idx] = 1;
completed++;
}
}
else {
current_time++;
}
}
int min_arrival_time = 10000000;
int max_completion_time = -1;
for(int i = 0; i < n; i++) {
min_arrival_time = min(min_arrival_time,p[i].arrival_time);
max_completion_time = max(max_completion_time,p[i].completion_time);
}
avg_turnaround_time = (float) total_turnaround_time / n;
avg_waiting_time = (float) total_waiting_time / n;
avg_response_time = (float) total_response_time / n;
cpu_utilisation = ((max_completion_time - total_idle_time) / (float) max_completion_time )*100;
throughput = float(n) / (max_completion_time - min_arrival_time);
cout<<endl<<endl;
cout<<"#P\t"<<"AT\t"<<"BT\t"<<"ST\t"<<"CT\t"<<"TAT\t"<<"WT\t"<<"RT\t"<<"\n"<<endl;
for(int i = 0; i < n; i++) {
cout<<p[i].pid<<"\t"<<p[i].arrival_time<<"\t"<<p[i].burst_time<<"\t"<<p[i].start_time<<"\t"<<p[i].completion_time<<"\t"<<p[i].turnaround_time<<"\t"<<p[i].waiting_time<<"\t"<<p[i].response_time<<"\t"<<"\n"<<endl;
}
cout<<"Average Turnaround Time = "<<avg_turnaround_time<<endl;
cout<<"Average Waiting Time = "<<avg_waiting_time<<endl;
cout<<"Average Response Time = "<<avg_response_time<<endl;
cout<<"CPU Utilization = "<<cpu_utilisation<<"%"<<endl;
cout<<"Throughput = "<<throughput<<" process/unit time"<<endl;
}
Leave a comment