-
Notifications
You must be signed in to change notification settings - Fork 1
/
JCTfromExec.py
212 lines (190 loc) · 6.88 KB
/
JCTfromExec.py
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
72
73
74
75
76
77
78
79
80
81
82
83
84
85
86
87
88
89
90
91
92
93
94
95
96
97
98
99
100
101
102
103
104
105
106
107
108
109
110
111
112
113
114
115
116
117
118
119
120
121
122
123
124
125
126
127
128
129
130
131
132
133
134
135
136
137
138
139
140
141
142
143
144
145
146
147
148
149
150
151
152
153
154
155
156
157
158
159
160
161
162
163
164
165
166
167
168
169
170
171
172
173
174
175
176
177
178
179
180
181
182
183
184
185
186
187
188
189
190
191
192
193
194
195
196
197
198
199
200
201
202
203
204
205
206
207
208
209
210
211
import sys,os,argparse
import random
import heapq
from numpy import mean
from copy import deepcopy
from itertools import groupby
from operator import itemgetter
from math import floor
def executionTime():
return random.expovariate(0.1)
def jobSizeGenerate():
return int(random.expovariate(0.015))
def mockPlace(heap, tasks_num,duration):
candidates=[]
for k in range(tasks_num):
(time,index) = heapq.heappop(heap)
candidates.append(index)
heapq.heappush(heap, (time+duration,index))
return candidates
def main(jobs_num, worker_num, probRatio=2):
stfSet=[]
srjfSet=[]
fifoSet=[]
taSet=[]
speedupOverSRJF=[]
speedupOverFIFO=[]
for iteration in range(20):
workers = [[] for i in range(worker_num)]
for jobIndex in range(jobs_num):
duration = executionTime()
tasks_num = max(jobSizeGenerate(),1)
probs = random.sample(range(worker_num), min(worker_num, tasks_num*probRatio))
#convert to list of waiting time of each worker
probs = map(lambda x: (sum([a for (a,b) in workers[x]]), x), probs)
heapq.heapify(probs)
candidates = mockPlace(probs,tasks_num,duration)
assert(len(candidates)>0)
for k in candidates:
workers[k].append((duration,jobIndex))
stf = STF(deepcopy(workers))
srjf = SRJF(deepcopy(workers))
fifo = FIFO(deepcopy(workers),2)
ta = tailAware(deepcopy(workers))
stfSet.append(sum(stf))
srjfSet.append(sum(srjf))
fifoSet.append(sum(fifo))
taSet.append(sum(ta))
#speedupOverSRJF.append(float(srjf)/ta)
#speedupOverFIFO.append(float(fifo)/ta)
print ("STF:", mean(stfSet))
print ("SRJF:", mean(srjfSet))
print ("FIFO:", mean(fifoSet))
print ("TailAware", mean(taSet))
print ("speedupOverSTF", mean(stfSet)/mean(taSet))
print ("speedupOverFIFO", mean(fifoSet)/mean(taSet))
#print ("len(ta)",len(ta),"len(fifo)",len(fifo),"len(srjf)",len(srjf),"len(workers)",\
# len(set([x for worker in workers for (y,x) in worker])))
#print ("speedupOverSRJF==================")
#print ("max",max(speedupOverSRJF),"min",min(speedupOverSRJF),"mean",mean(speedupOverSRJF))
#print ("speedupOverFIFO==================")
#print ("max",max(speedupOverFIFO),"min",min(speedupOverFIFO),"mean",mean(speedupOverFIFO))
# print "srjf================"
# print srjf
# print "fifo================"
# print fifo
# print "ta=================="
# print ta
def SRJF(placements):
#placements is a list of list of tasks(duration, jobIndex) on a worker
execLog = []
startTimeLog = {}
timeAccu =[0]*len(placements)
jobOrder =[]
totalRemainTasks = sum([len(x) for x in placements])
JobRemTracker = {}
def tallyEachJob(placements):
items = [item for sublist in placements for item in sublist]
items.sort(key=itemgetter(1))
remainWork = [reduce(lambda x,y: (x[0]+y[0],x[1]),group) \
for _,group in groupby(items, key=itemgetter(1))]
remainWork.sort(key=itemgetter(0))
return dict([(y,x) for (x,y) in remainWork])
def findClosestKey(JobTracker, key):
assert(key >=0)
keys = JobTracker.keys()
keys.sort(reverse=True)
for k in keys:
if k <= key:
return k
JobRemTracker[0] = tallyEachJob(placements)
while(totalRemainTasks > 0):
for i in range(len(placements)):
if totalRemainTasks > 0:
mostRecentKey = findClosestKey(JobRemTracker,timeAccu[i])
jobOrder = JobRemTracker[mostRecentKey]
if(len(placements[i]) > 0):
(duration, jobIndex) = min(placements[i], key=lambda x: jobOrder[x[1]] )
placements[i].remove((duration,jobIndex))
if (jobIndex not in startTimeLog.keys() or startTimeLog[jobIndex]> timeAccu[i]):
startTimeLog[jobIndex] = timeAccu[i]
JobRemTracker[timeAccu[i]]=tallyEachJob(placements)
execLog.append((timeAccu[i]+duration,jobIndex))
timeAccu[i] += duration
totalRemainTasks -= 1
execLog.sort(key=itemgetter(1))
JCT = [reduce(lambda x,y: (max(x[0],y[0]),x[1]), group) for _,group in groupby(execLog,key=itemgetter(1))]
return [ x- startTimeLog[y] for (x,y) in JCT]
def STF(placements):
execLog=[]
startTimeLog= {}
for li in placements:
li.sort(key=itemgetter(0))
for li in placements:
acc = 0
#print li
for (duration, jobIndex) in li:
if(jobIndex not in startTimeLog.keys() or acc < startTimeLog[jobIndex]):
startTimeLog[jobIndex] = acc
execLog.append( (duration+acc, jobIndex) )
acc += duration
execLog.sort(key=itemgetter(1))
JCT = [reduce(lambda x,y: (max(x[0],y[0]),x[1]), group) for _,group in groupby(execLog,key=itemgetter(1))]
return [x - startTimeLog[y] for (x,y) in JCT]
def FIFO(placements,flag):
execLog=[]
startTimeLog={}
for li in placements:
acc = 0
# if(flag>=2):
# #print li
for (duration, jobIndex) in li:
if(jobIndex not in startTimeLog.keys() or acc < startTimeLog[jobIndex]):
startTimeLog[jobIndex] = acc
execLog.append( (duration+acc, jobIndex) )
acc += duration
execLog.sort(key=itemgetter(1))
JCT = [reduce(lambda x,y: (max(x[0],y[0]),x[1]), group) for _,group in groupby(execLog,key=itemgetter(1))]
if(flag>=1):
return [x - startTimeLog[y] for (x,y) in JCT]
else:
return dict([(y,x) for (x,y) in JCT])
def tailAware(placements):
def checkAndPerform(worker,budgetsPerWorker,indexOfBottleneck):
cur = indexOfBottleneck
if(cur <= 0):
return
myJobIndex = worker[cur][1]
leftJobIndex = worker[cur-1][1]
if worker[cur-1][0] > worker[cur][0] \
and budgetsPerWorker[leftJobIndex] >= worker[cur][0]:
budgetsPerWorker[myJobIndex] += worker[cur-1][0]
budgetsPerWorker[leftJobIndex] -= worker[cur][0]
temp = worker[cur]
worker[cur] = worker[cur-1]
worker[cur-1] = temp
checkAndPerform(worker, budgetsPerWorker, indexOfBottleneck-1)
#merge tasks of the same job on every worker
for li in placements:
if(len(li) > 1):
k = 1
while(k < len(li)):
if(li[k][1] == li[k-1][1]):
li[k-1] = (li[k-1][0]+li[k][0], li[k-1][1])
li.pop(k)
else:
k += 1
tailLatency = FIFO(deepcopy(placements),0)
budgets = [{} for i in range(len(placements))]
#update budgets for each task on each worker
for ii in range(len(placements)):
acc = 0
for (duration, jobIndex) in placements[ii]:
budgets[ii][jobIndex] = tailLatency[jobIndex]-(acc+duration)
assert(budgets[ii][jobIndex]>=0)
acc += duration
for ii in range(len(placements)):
#print placements[ii]
for jj in range(len(placements[ii])):
checkAndPerform(placements[ii], budgets[ii], jj)
return FIFO(deepcopy(placements),1)
if __name__ == "__main__":
parser = argparse.ArgumentParser()
parser.add_argument("jobNum", type=int, help= "specify how many jobs")
parser.add_argument("workerNum", type=int, help= "how many workers")
#parser.add_argument("tasksNum", type=int, help = "how many tasks in a job")
#parser.add_argument("taskHeterogeneity", type=int, help="ratio of the longest task over the shrotest task")
#parser.add_argument("probRatio", type=int, help = "how many probs for each task",default=2)
#parser.add_argument("iterations", type=int, help="specify the number of iterations",default=1)
args = parser.parse_args()
main(args.jobNum,args.workerNum)