利用python画出SJF调度图
最先发布在csdn。本人原创。
https://blog.csdn.net/weixin_43906799/article/details/105510046
SJF算法:
最短作业优先(SJF)调度算法将每个进程与其下次 CPU 执行的长度关联起来。实际上,短进程/作业(要求服务时间最短)在实际情况中占有很大比例,为了使得它们优先执行,追求最少的平均等待时间时间、平均周转时间、平均带权周转时间。短作业优先可能导致长作业一直得不到处理)
总体构想
用python绘图这个想法产生于写调度图作业那段时间。当时就想着用python绘图,有两个想法trutle动态绘制调度图,还有就是现在所使用的方法。为什么用类写这次的作业,一是下次的作业可以直接继承SJF类,然后修改调度函数和排序函数就行了。二是用类写代码解决一类问题,代码看起来比较漂亮。
算法设计结构图
程序执行结果图
作业信息
作业名 | 到达时间 | 运行时间 |
---|---|---|
A | 0 | 5 |
B | 1 | 4 |
C | 2 | 1 |
D | 4 | 2 |
E | 5 | 1 |
基本思路
(1)类初始化:
对于进程调度SJF算法这个类,首先我们需要有成员变量,也就是大致所需要的成员变量。 基本也就需要这么多。
self.data = [] | 存储进程 |
---|---|
self.name = '' | 进程名字 |
self.service_time = 0 | 服务时间 |
self.arrival_time = 0 | 到达时间 |
self.state = '' | 初始状态 |
self.number = 0 | 进程数量 |
self.timeout = 0 | 超时限定 |
self.start = 0 | 开始时间 |
self.end = 0 | 结束时间 |
def __init__(self):
super(Solution, self).__init__()
# save tasks
self.data = []
self.name = ''
self.service_time = 0
self.arrival_time = 0
self.state = ''
self.number = 0
self.timeout = 0
self.start = 0
self.end = 0
(2)获取数据:
获取数据可以从文件(如.txt)中读入,亦可以从console读入。这里要求一个地方,就是数据的格式,名字,到达时间,服务时间。中间用空格分开。如下面表格:
name | arrival_time | service_time |
---|---|---|
A | 0 | 5 |
B | 1 | 4 |
C | 2 | 1 |
D | 4 | 2 |
E | 5 | 1 |
def get_data_file(self):
with open('data.txt', "r", encoding="utf-8") as file:
for line in file.read().splitlines():
name, arrival_time, service_time = line.split()
# insert the task
self.insert_data(name, arrival_time, service_time)
file.close()
# initial queue
# sort first arrival_time and second service_time
self.data.sort(key=lambda x: (x['arrival_time'], x['service_time']))
# update and recode id
for i in range(self.number):
self.data[i]['index'] = i
def get_data_input(self):
print('How many tasks do you want input?')
tasks_number = int(input('Please enter an integer of type int:'))
print('Please enter name and arrival_time and service_time of task')
print('such as:A 0 5')
for _ in range(tasks_number):
name, arrival_time, service_time = input('Please enter\n').split()
self.insert_data(name, arrival_time, service_time)
# initial queue
# sort first arrival_time and second service_time
self.data.sort(key=lambda x: (x['arrival_time'], x['service_time']))
# update and recode id
for i in range(self.number):
self.data[i]['index'] = i
(3)进行调度:
也就是设计算法,来实现SJF。基本的算法思路,就是维护一个优先队列。如图:
每次调度的时候根据需要,然后更新信息,更改作业的状态和到达和结束的时间。同时获取下一个或者多个作业,这里需要考虑到一种情况,就是当前时间片不能获取下一个作业,需要等待一段时间作业到达,才能执行。这种情况特判一下。然后执行排序,维护这个优先队列。
def implement(self):
'''start algorithm'''
# get first task
data = [self.data[0]]
# update the time of start
self.start = self.end = data[0]['arrival_time']
while data:
# update information
self.update_information(
data[0]['index'], self.end, self.end + data[0]['service_time'])
# get next task or tasks
data += self.get_next_data(data.pop(0)['index'], data)
# maintain the queue
data = self.sort_data(data)
self.data.sort(key=lambda x: x['id'])
(4)排序和信息更新:
对于排序的实现其实很简单,前面的结构图也已经展示了,对于SJF算法一共有两种排序方式,分别在不同的过程进行使用。数据更新就是更新原始的数据,包括计算状态,开始时间,结束时间,周转时间,平均周转时间等等。
def update_information(self, index, start, end):
self.data[index]['start'] = start
self.data[index]['end'] = end
self.data[index]['state'] = 'f'
self.data[index]['turnaround_time'] = end - \
self.data[index]['arrival_time']
self.data[index]['authorized_turnover_time'] = self.data[index]['turnaround_time'] / \
self.data[index]['service_time']
self.start = start
self.end = end
self.show_data_running(start, end, self.data[index])
(5)数据输出:
为什么要数据输出,其实这就是一个数据可视化的一种方法。也就是直观的表达各种信息。所以数据输出部分,就是自己设置自己的排版,布局,可以利用\t制表符来打表。
def show_data(self):
print("{:<6}{:<10}{:<10}{:<10}{:<6}{:<8}{:<7}{:<6}".format(
'name', 'arr_time', 'ser_time', 'state', '周转时间', '带权周转时间', 'start', 'end'))
for task in sorted(self.data, key=lambda x: x['id']):
print("{:<6}{:<10}{:<10}{:<10}{:<10}{:<14.2f}{:<7}{:<4}".format(
task['name'],
task['arrival_time'],
task['service_time'],
task['state'],
task['turnaround_time'],
task['authorized_turnover_time'],
task['start'],
task['end']))
(6)plt生成调度图展示:
利用python的第三方库,根据数据进行绘图,然后展示出好看的图片。
def init_image(self):
# size = 1000 * 500
plt.figure('SJF', figsize=(10, 5))
self.drow_image()
# setting xticks for 0 to self.end + 2
plt.xticks([i for i in range(self.end + 3)])
# setting title
plt.title('the time of task about SJF')
plt.xlabel('')
plt.ylabel('tasks')
# setting yticks.such as A == 0
plt.yticks(self.get_y_ticks()[0], self.get_y_ticks()[1])
def drow_image(self):
for task in self.data:
# the time line of task from start to end
plt.plot([task['start'], task['end']],
[task['id'], task['id']],
label=task['name'],
lw=2)
# annotation of the key point
plt.plot([task['end'], task['end']],
[-1, task['id']],
'k--',
lw=1)
# legend
plt.legend(loc='best')
def set_ax(self):
ax = plt.gca() # 获取到当前坐标轴信息
ax.spines['right'].set_color('none')
ax.spines['bottom'].set_color('none')
ax.xaxis.set_ticks_position('top') # 将X坐标轴移到上面
ax.invert_yaxis() # 反转Y坐标轴
ax.grid(True, linestyle='-.') # 网格
def show_image(self):
self.init_image()
self.set_ax()
plt.savefig('SJF.png', dpi=300)
plt.show()
程序执行过程:
支持两种输入方式,手动输入和数据导入。
数据导入:
原始数据
调度前:
调度中:
调度后:
生成调度图:
手动输入数据:
调度前
调度中
调度后
生成调度图:
程序源代码:
# -*- coding: utf-8 -*-
# @Author: wfy
# @Date: 2020-04-10 15:31:44
# @Last Modified by: wfy
# @Last Modified time: 2020-04-14 13:46:31
import matplotlib.pyplot as plt
class Solution():
"""to achieve SJF"""
def __init__(self):
super(Solution, self).__init__()
# save tasks
self.data = []
self.name = ''
self.service_time = 0
self.arrival_time = 0
self.state = ''
self.number = 0
self.timeout = 0
self.start = 0
self.end = 0
def insert_data(self, name, arrival_time, service_time):
self.data.append({
'id': self.number,
'name': name,
'arrival_time': int(arrival_time),
'service_time': int(service_time),
'state': 'w',
'turnaround_time': 0,
'authorized_turnover_time': 0,
'start': 0,
'end': 0
})
self.timeout = max(self.timeout, int(arrival_time))
self.number += 1
def get_data_file(self):
with open('data.txt', "r", encoding="utf-8") as file:
for line in file.read().splitlines():
name, arrival_time, service_time = line.split()
# insert the task
self.insert_data(name, arrival_time, service_time)
file.close()
# initial queue
# sort first arrival_time and second service_time
self.data.sort(key=lambda x: (x['arrival_time'], x['service_time']))
# update and recode id
for i in range(self.number):
self.data[i]['index'] = i
def get_data_input(self):
print('How many tasks do you want input?')
tasks_number = int(input('Please enter an integer of type int:'))
print('Please enter name and arrival_time and service_time of task')
print('such as:A 0 5')
for _ in range(tasks_number):
name, arrival_time, service_time = input('Please enter\n').split()
self.insert_data(name, arrival_time, service_time)
# initial queue
# sort first arrival_time and second service_time
self.data.sort(key=lambda x: (x['arrival_time'], x['service_time']))
# update and recode id
for i in range(self.number):
self.data[i]['index'] = i
def show_data_running(self, start, end, data):
print('-'*40)
print("from {:} to {:}".format(start, end))
print("task name:{:}".format(data['name']))
print("task state:{:}\n".format('R'))
def show_data(self):
print("{:<6}{:<10}{:<10}{:<10}{:<6}{:<8}{:<7}{:<6}".format(
'name', 'arr_time', 'ser_time', 'state', '周转时间', '带权周转时间', 'start', 'end'))
for task in sorted(self.data, key=lambda x: x['id']):
print("{:<6}{:<10}{:<10}{:<10}{:<10}{:<14.2f}{:<7}{:<4}".format(
task['name'],
task['arrival_time'],
task['service_time'],
task['state'],
task['turnaround_time'],
task['authorized_turnover_time'],
task['start'],
task['end']))
def cmp(self):
'''the method of sort'''
return lambda x: (x['service_time'], x['arrival_time'], x['index'])
def sort_data(self, data):
return sorted(data, key=self.cmp())
def update_information(self, index, start, end):
self.data[index]['start'] = start
self.data[index]['end'] = end
self.data[index]['state'] = 'f'
self.data[index]['turnaround_time'] = end - \
self.data[index]['arrival_time']
self.data[index]['authorized_turnover_time'] = self.data[index]['turnaround_time'] / \
self.data[index]['service_time']
self.start = start
self.end = end
self.show_data_running(start, end, self.data[index])
def get_next_data(self, index, data):
# get tasks from the beginning to the end of the current task
result = [x for x in self.data if x['arrival_time'] <=
self.end and x['state'] == 'w' and x not in data]
if result or data:
return result
# no tasks entered at current time
for task in self.data:
if task['state'] == 'w':
self.start = self.end = task['arrival_time']
return [task]
return []
def implement(self):
'''start algorithm'''
# get first task
data = [self.data[0]]
# update the time of start
self.start = self.end = data[0]['arrival_time']
while data:
# update information
self.update_information(
data[0]['index'], self.end, self.end + data[0]['service_time'])
# get next task or tasks
data += self.get_next_data(data.pop(0)['index'], data)
# maintain the queue
data = self.sort_data(data)
self.data.sort(key=lambda x: x['id'])
def get_y_ticks(self):
return [x['id'] for x in self.data] + [self.data[-1]['id'] + 1], [x['name'] for x in self.data] + ['']
def init_image(self):
# size = 1000 * 500
plt.figure('SJF', figsize=(10, 5))
self.drow_image()
# setting xticks for 0 to self.end + 2
plt.xticks([i for i in range(self.end + 3)])
# setting title
plt.title('the time of task about SJF')
plt.xlabel('')
plt.ylabel('tasks')
# setting yticks.such as A == 0
plt.yticks(self.get_y_ticks()[0], self.get_y_ticks()[1])
def drow_image(self):
for task in self.data:
# the time line of task from start to end
plt.plot([task['start'], task['end']],
[task['id'], task['id']],
label=task['name'],
lw=2)
# annotation of the key point
plt.plot([task['end'], task['end']],
[-1, task['id']],
'k--',
lw=1)
# legend
plt.legend(loc='best')
def set_ax(self):
ax = plt.gca() # 获取到当前坐标轴信息
ax.spines['right'].set_color('none')
ax.spines['bottom'].set_color('none')
ax.xaxis.set_ticks_position('top') # 将X坐标轴移到上面
ax.invert_yaxis() # 反转Y坐标轴
ax.grid(True, linestyle='-.') # 网格
def show_image(self):
self.init_image()
self.set_ax()
plt.savefig('SJF.png', dpi=300)
plt.show()
def main(self):
if input('Do you want get data by file? y/Y or n/N\n') in ['y', 'Y']:
SJF.get_data_file()
else:
SJF.get_data_input()
SJF.show_data()
SJF.implement()
SJF.show_data()
SJF.show_image()
if __name__ == '__main__':
try:
SJF = Solution()
SJF.main()
except Exception as e:
print('An exception', e)
else:
print('Finish')
finally:
print('finally')