A Python port of the Workload Model proposed by Lublin & Feitelson.
The following code shows how to use it:
from parallelworkloads import lublin99
w = lublin99.Lublin99(1, 2) # Will use both batch and interactive jobs
w.numJobs=4 # will generate four jobs
w.generate() # The four generated jobs are shown below
[SwfJob(jobId=1, submissionTime=103, waitTime=-1, runTime=12379,
allocProcs=16, avgCpuUsage=-1, usedMem=-1, reqProcs=-1, reqTime=-1,
reqMem=-1, status=1, userId=-1, groupId=-1, executable=-1, queueNum=1,
partNum=-1, precedingJob=-1, thinkTime=-1),
SwfJob(jobId=2, submissionTime=3089, waitTime=-1, runTime=177,
allocProcs=16, avgCpuUsage=-1, usedMem=-1, reqProcs=-1, reqTime=-1,
reqMem=-1, status=1, userId=-1, groupId=-1, executable=-1, queueNum=1,
partNum=-1, precedingJob=-1, thinkTime=-1),
SwfJob(jobId=3, submissionTime=3150, waitTime=-1, runTime=10, allocProcs=2,
avgCpuUsage=-1, usedMem=-1, reqProcs=-1, reqTime=-1, reqMem=-1, status=1,
userId=-1, groupId=-1, executable=-1, queueNum=0, partNum=-1,
precedingJob=-1, thinkTime=-1),
SwfJob(jobId=4, submissionTime=3172, waitTime=-1, runTime=7,
allocProcs=32, avgCpuUsage=-1, usedMem=-1, reqProcs=-1, reqTime=-1,
reqMem=-1, status=1, userId=-1, groupId=-1, executable=-1, queueNum=0,
partNum=-1, precedingJob=-1, thinkTime=-1)]parallelworkloads also supports generating runtime estimates based on the model proposed by Dan Tsafrir in 2005. For the model to work, it needs at least 200 jobs. Here's an example continuing the previous one:
from parallelworkloads import tsafrir05
w.numJobs = 200
jobs = w.generate()
print('Original requested time of first job:', jobs[0].reqTime)
t = tsafrir05.Tsafrir05(jobs)
jobs = t.generate(jobs)
print('Generated requested time of first job:', jobs[0].reqTime)Which gives as output:
Original requested time: -1 Generated requested time: 22962.0