![Page 1: Uwe Schwiegelshohn EPIT 2007, June 5 Ordonnancementvirtualpanic.com/anonymousftplistings/ebooks/Job Scheduling... · Uwe Schwiegelshohn EPIT 2007, June 5 Ordonnancement. 2 Content](https://reader031.vdocuments.net/reader031/viewer/2022030207/5b9d0ad109d3f2de128b7c6d/html5/thumbnails/1.jpg)
University DortmundRobotics Research InstituteInformation Technology
Job Scheduling
Uwe Schwiegelshohn
EPIT 2007, June 5Ordonnancement
![Page 2: Uwe Schwiegelshohn EPIT 2007, June 5 Ordonnancementvirtualpanic.com/anonymousftplistings/ebooks/Job Scheduling... · Uwe Schwiegelshohn EPIT 2007, June 5 Ordonnancement. 2 Content](https://reader031.vdocuments.net/reader031/viewer/2022030207/5b9d0ad109d3f2de128b7c6d/html5/thumbnails/2.jpg)
2
Content of the Lecture
What is job scheduling?
Single machine problems and results
Makespan problems on parallel machines
Utilization problems on parallel machines
Completion time problems on parallel machines
Exemplary workload problem
![Page 3: Uwe Schwiegelshohn EPIT 2007, June 5 Ordonnancementvirtualpanic.com/anonymousftplistings/ebooks/Job Scheduling... · Uwe Schwiegelshohn EPIT 2007, June 5 Ordonnancement. 2 Content](https://reader031.vdocuments.net/reader031/viewer/2022030207/5b9d0ad109d3f2de128b7c6d/html5/thumbnails/3.jpg)
3
Examples of Job Scheduling
Processor schedulingJobs are executed on a CPU in a multitasking operating system. Users submit jobs to web servers and receive results after some time.Users submit batch computing jobs to a parallel processor.
Bandwidth schedulingUsers call other persons and need bandwidth for some period of time.
Airport gate schedulingAirlines require gates for their flights at an airport.
Repair crew schedulingCustomer request the repair of their devices.
![Page 4: Uwe Schwiegelshohn EPIT 2007, June 5 Ordonnancementvirtualpanic.com/anonymousftplistings/ebooks/Job Scheduling... · Uwe Schwiegelshohn EPIT 2007, June 5 Ordonnancement. 2 Content](https://reader031.vdocuments.net/reader031/viewer/2022030207/5b9d0ad109d3f2de128b7c6d/html5/thumbnails/4.jpg)
4
Job Properties
Independent jobsNo known precedence constraints
Difference to task scheduling
Atomic jobsNo job stages
Difference to job shop scheduling
Batch jobsNo deadlines or due dates
Difference to deadline scheduling
pj processing time of job jrj release date of job j earliest starting time
importance of the jobparallelism of the job
wj weight of job jmj size of job j
![Page 5: Uwe Schwiegelshohn EPIT 2007, June 5 Ordonnancementvirtualpanic.com/anonymousftplistings/ebooks/Job Scheduling... · Uwe Schwiegelshohn EPIT 2007, June 5 Ordonnancement. 2 Content](https://reader031.vdocuments.net/reader031/viewer/2022030207/5b9d0ad109d3f2de128b7c6d/html5/thumbnails/5.jpg)
5
Machine Environments
1: single machine Many job scheduling problems are easy.
Pm: m parallel identical machinesEvery job requires the same processing time on each machine.Use of machine eligibility constraints Mj if job j can only be executed on a subset of machines
Airport gate scheduling: wide and narrow body airplanes
Qm: m uniformly related machinesThe machines have different speeds vi that are valid for all jobs.In deterministic scheduling, results for Pm and Qm are related.In online scheduling, there are significant differences between Pmand Qm.
Rm: m unrelated machinesEach job has a different processing time on each machine.
![Page 6: Uwe Schwiegelshohn EPIT 2007, June 5 Ordonnancementvirtualpanic.com/anonymousftplistings/ebooks/Job Scheduling... · Uwe Schwiegelshohn EPIT 2007, June 5 Ordonnancement. 2 Content](https://reader031.vdocuments.net/reader031/viewer/2022030207/5b9d0ad109d3f2de128b7c6d/html5/thumbnails/6.jpg)
6
Restrictions and Constraints
Release dates rjParallelism mj
Fixed parallelism: mj machines must be available during the whole processing of the job.Malleable jobs: The number of allocated machines can change before or during the processing of the job.
PreemptionThe processing of a job can be interrupted and continued on another machine.Gang scheduling: The processing of a job must be continued on the same machines.
Machine eligibility constraints MjBreakdown of machines
m(t): time dependent availability
rarely discussed in the literature
![Page 7: Uwe Schwiegelshohn EPIT 2007, June 5 Ordonnancementvirtualpanic.com/anonymousftplistings/ebooks/Job Scheduling... · Uwe Schwiegelshohn EPIT 2007, June 5 Ordonnancement. 2 Content](https://reader031.vdocuments.net/reader031/viewer/2022030207/5b9d0ad109d3f2de128b7c6d/html5/thumbnails/7.jpg)
7
Objective Functions
Completion time of job j: Cj
Owner oriented:Makespan: Cmax =max (C1 ,...,Cn )
completion time of the last job in the systemUtilization Ut: Average ratio of busy machines to all machines in the interval (0,t] for some time t.
User oriented:Total completion time: Σ Cj
Total weighted completion time: Σ wj Cj
Total weighted waiting time: Σ wj ( Cj –pj – rj ) = Σ wj Cj – Σ wj (pj+rj)Total weighted flow time: Σ wj ( Cj – rj ) = Σ wj Cj – Σ wj rj
Regular objective functions:non decreasing in C1 ,...,Cn
const.const.
![Page 8: Uwe Schwiegelshohn EPIT 2007, June 5 Ordonnancementvirtualpanic.com/anonymousftplistings/ebooks/Job Scheduling... · Uwe Schwiegelshohn EPIT 2007, June 5 Ordonnancement. 2 Content](https://reader031.vdocuments.net/reader031/viewer/2022030207/5b9d0ad109d3f2de128b7c6d/html5/thumbnails/8.jpg)
8
Workload Classification
Deterministic scheduling problemsAll problem parameters are available at time 0.Optimal algorithms, Simple individual approximation algorithmsPolynomial time approximation schemes
Online scheduling problemsParameters of job j are unknown until rj (submission over time).pj is unknown Cj (nonclairvoyant scheduling).Competitive analysis
Stochastic schedulingKnown distribution of job parametersRandomized algorithms
Workload based schedulingAn algorithm is parameterized to achieve a good solution for a given workload.
![Page 9: Uwe Schwiegelshohn EPIT 2007, June 5 Ordonnancementvirtualpanic.com/anonymousftplistings/ebooks/Job Scheduling... · Uwe Schwiegelshohn EPIT 2007, June 5 Ordonnancement. 2 Content](https://reader031.vdocuments.net/reader031/viewer/2022030207/5b9d0ad109d3f2de128b7c6d/html5/thumbnails/9.jpg)
9
No machine is kept idle while a job is waiting for processing.An optimal schedule need not be nondelay!
Example: 1 | | Σ wj Cj
jobs 1 2pj 1 3rj 1 0wj 2 1
0 5
12 Σ wj Cj=11
Nondelay schedule
1 2 Σ wj Cj=9
Optimal schedule
Nondelay (Greedy) Schedule
![Page 10: Uwe Schwiegelshohn EPIT 2007, June 5 Ordonnancementvirtualpanic.com/anonymousftplistings/ebooks/Job Scheduling... · Uwe Schwiegelshohn EPIT 2007, June 5 Ordonnancement. 2 Content](https://reader031.vdocuments.net/reader031/viewer/2022030207/5b9d0ad109d3f2de128b7c6d/html5/thumbnails/10.jpg)
10
Complexity Hierarchy
Some problems are special cases of other problems:Notation: α1 | β1 | γ1 ∝ (reduces to) α2 | β2 | γ2
Examples:1 || Σ Cj ∝ 1 || Σ wj Cj ∝ Pm || Σ wj Cj ∝ Pm | mj | Σ wj Cj
prmp
0Pm
1
brkdwn
0
Mj
0
mj
1
wj
1
rj
0
Σwj Cj
ΣCj
Rm
Qm
![Page 11: Uwe Schwiegelshohn EPIT 2007, June 5 Ordonnancementvirtualpanic.com/anonymousftplistings/ebooks/Job Scheduling... · Uwe Schwiegelshohn EPIT 2007, June 5 Ordonnancement. 2 Content](https://reader031.vdocuments.net/reader031/viewer/2022030207/5b9d0ad109d3f2de128b7c6d/html5/thumbnails/11.jpg)
11
Content of the Lecture
What is job scheduling?
Single machine problems and results
Makespan problems on parallel machines
Utilization problems on parallel machines
Completion time problems on parallel machines
Exemplary workload problem
![Page 12: Uwe Schwiegelshohn EPIT 2007, June 5 Ordonnancementvirtualpanic.com/anonymousftplistings/ebooks/Job Scheduling... · Uwe Schwiegelshohn EPIT 2007, June 5 Ordonnancement. 2 Content](https://reader031.vdocuments.net/reader031/viewer/2022030207/5b9d0ad109d3f2de128b7c6d/html5/thumbnails/12.jpg)
12
1 || Σ wj Cj
1 || Σ wj Cj is easy and can be solved by sorting all jobs in decreasing Smith order wj/pj (weighted shortest processing time first (WSPT) rule, Smith, 1956).
Nondelay scheduleProof by contradiction and localization: If the WSPT rule is violated then it is violated by a pair of neighboring task h and k.
S1: Σ wj Cj = ...+ wh(t+ph) + wk(t + ph + pk)
ht
k
S2: Σ wj Cj = ...+ wk(t+pk) + wh(t + pk + ph)
k h
S1-S2: wk ph – wh pk > 0
wk/pk > wh/ph
![Page 13: Uwe Schwiegelshohn EPIT 2007, June 5 Ordonnancementvirtualpanic.com/anonymousftplistings/ebooks/Job Scheduling... · Uwe Schwiegelshohn EPIT 2007, June 5 Ordonnancement. 2 Content](https://reader031.vdocuments.net/reader031/viewer/2022030207/5b9d0ad109d3f2de128b7c6d/html5/thumbnails/13.jpg)
13
Other Single Machine Problems
Every nondelay schedule has optimal makespan and optimal utilization for any interval starting at time 0.
WSPT requires knowledge of the processing timesNo direct application to nonclairvoyant scheduling
1 | prmp | Σ Cj is easy.The online nonclairvoyant version (Round Robin) has a competitive factor of 2-2/(n+1) (Motwani, Phillips, Torng,1994).
1 | rj ,prmp | Σ Cj is easy. The online, clairvoyant version is easy.
1 | rj | Σ Cj is strongly NP hard.1 | rj ,prmp | Σ wj Cj is strongly NP hard.
The WSRPT (remaining processing time) rule is not optimal.
![Page 14: Uwe Schwiegelshohn EPIT 2007, June 5 Ordonnancementvirtualpanic.com/anonymousftplistings/ebooks/Job Scheduling... · Uwe Schwiegelshohn EPIT 2007, June 5 Ordonnancement. 2 Content](https://reader031.vdocuments.net/reader031/viewer/2022030207/5b9d0ad109d3f2de128b7c6d/html5/thumbnails/14.jpg)
14
Optimal versus Approximation
1 | rj ,prmp | Σ wj (Cj-rj) and 1 | rj ,prmp | Σ wj Cj
Same optimal solutionLarger approximation factor for 1 | rj ,prmp | Σ wj (Cj-rj).No constant approximation factor for the total flowtime objective (Kellerer, Tautenhahn, Wöginger, 1999)
∑∑
∑∑
∑∑
−⋅
⋅⋅⎟⎟⎠
⎞⎜⎜⎝
⎛−
⋅
⋅+
⋅
⋅=
)rOPT)(C(wrw
1OPT)(Cw
S)(CwOPT)(Cw
S)(Cw
jjj
jj
jj
jj
jj
jj
=−⋅
−⋅
∑∑
)r OPT)(C(w)r(S)C(w
jjj
jjj
=−⋅
⋅⎟⎟⎠
⎞⎜⎜⎝
⎛−
⋅⋅
+−⋅⋅⋅
=∑
∑ ∑∑∑
∑∑
)rOPT)(C(w
rw1OPT)(Cw
S)(Cw)rOPT)(C(w
OPT)(CwS)(Cw
jjj
jjjj
jjjjj
jj
jj
![Page 15: Uwe Schwiegelshohn EPIT 2007, June 5 Ordonnancementvirtualpanic.com/anonymousftplistings/ebooks/Job Scheduling... · Uwe Schwiegelshohn EPIT 2007, June 5 Ordonnancement. 2 Content](https://reader031.vdocuments.net/reader031/viewer/2022030207/5b9d0ad109d3f2de128b7c6d/html5/thumbnails/15.jpg)
15
Approximation Algorithms
1 | rj | Σ CjApproximation factor e/(e-1)=1.58 (Chekuri, Motwani, Natarajan, Stein, 2001)Clairvoyant online scheduling: competitive factor 2 (Hoogeveen, Vestjens,1996)
1 | rj | Σ wjCjApproximation factor 1.6853 (Goemans, Queyranne, Schulz, Skutella, Wang, 2002)Clairvoyant online scheduling: competitive factor 2 (Anderson, Potts, 2004)
1 | rj ,prmp | Σ wj CjApproximation factor 1.3333, Randomized online algorithm with the competitive factor 1.3333WSPT online algorithm with competitive factor 2 (all results: Schulz, Skutella, 2002)
![Page 16: Uwe Schwiegelshohn EPIT 2007, June 5 Ordonnancementvirtualpanic.com/anonymousftplistings/ebooks/Job Scheduling... · Uwe Schwiegelshohn EPIT 2007, June 5 Ordonnancement. 2 Content](https://reader031.vdocuments.net/reader031/viewer/2022030207/5b9d0ad109d3f2de128b7c6d/html5/thumbnails/16.jpg)
16
Content of the Lecture
What is job scheduling?
Single machine problems and results
Makespan problems on parallel machines
Utilization problems on parallel machines
Completion time problems on parallel machines
Exemplary workload problem
![Page 17: Uwe Schwiegelshohn EPIT 2007, June 5 Ordonnancementvirtualpanic.com/anonymousftplistings/ebooks/Job Scheduling... · Uwe Schwiegelshohn EPIT 2007, June 5 Ordonnancement. 2 Content](https://reader031.vdocuments.net/reader031/viewer/2022030207/5b9d0ad109d3f2de128b7c6d/html5/thumbnails/17.jpg)
17
Pm and Makespan with mj=1
A scheduling problem for parallel machines consists of 2 steps:
Allocation of jobs to machinesGenerating a sequence of the jobs on a machine
A minimal makespan represents a balanced load on the machines if no single job dominates the schedule.
Preemption may improve a schedule even if all jobs are released at the same time.
Optimal schedules for parallel identical machines are nondelay.
{ }⎭⎬⎫
⎩⎨⎧ ⋅≥ ∑ jjmax p
m1,p maxmaxOPT)(C
{ }⎭⎬⎫
⎩⎨⎧ ⋅= ∑ jjmax p
m1,p maxmaxOPT)(C
![Page 18: Uwe Schwiegelshohn EPIT 2007, June 5 Ordonnancementvirtualpanic.com/anonymousftplistings/ebooks/Job Scheduling... · Uwe Schwiegelshohn EPIT 2007, June 5 Ordonnancement. 2 Content](https://reader031.vdocuments.net/reader031/viewer/2022030207/5b9d0ad109d3f2de128b7c6d/html5/thumbnails/18.jpg)
18
Pm || Cmax
Pm || Cmax is strongly NP-hard (Garey, Johnson 1979).Approximation algorithm: Longest processing time first (LPT) rule (Graham, 1966)
Whenever a machine is free, the longest job among those not yet processed is put on this machine.
Tight approximation factor:
The optimal schedule Cmax(OPT) is not necessarily known but a simple lower bound can be used:
3m1
34
(OPT)C(LPT)C
max
max −≤
∑=
≥n
1jjmax p
m1(OPT)C
![Page 19: Uwe Schwiegelshohn EPIT 2007, June 5 Ordonnancementvirtualpanic.com/anonymousftplistings/ebooks/Job Scheduling... · Uwe Schwiegelshohn EPIT 2007, June 5 Ordonnancement. 2 Content](https://reader031.vdocuments.net/reader031/viewer/2022030207/5b9d0ad109d3f2de128b7c6d/html5/thumbnails/19.jpg)
19
LPT Proof (1)
If the claim is not true, then there is a counterexample with the smallest number n of jobs.The shortest job n in this counterexample is the last job to start processing (LPT) and the last job to finish processing.
If n is not the last job to finish processing then deletion of n does not change Cmax (LPT) while Cmax (OPT) cannot increase.A counter example with n – 1 jobs
Under LPT, job n starts at time Cmax(LPT)-pn.In time interval [0, Cmax(LPT) – pn], all machines are busy.
∑−
=
≤−1n
1jjnmax p
m1p(LPT)C
![Page 20: Uwe Schwiegelshohn EPIT 2007, June 5 Ordonnancementvirtualpanic.com/anonymousftplistings/ebooks/Job Scheduling... · Uwe Schwiegelshohn EPIT 2007, June 5 Ordonnancement. 2 Content](https://reader031.vdocuments.net/reader031/viewer/2022030207/5b9d0ad109d3f2de128b7c6d/html5/thumbnails/20.jpg)
20
LPT Proof (2)
1(OPT)C
)m1(1p(OPT)C
pm1
(OPT)C
)m1(1p
(OPT)C(LPT)C
3m1
34
max
n
max
n
1jj
max
n
max
max +−
≤+−
≤<−∑=
∑∑=
−
=
+−=+≤n
1jjn
1n
1jjnmax p
m1)
m1(1pp
m1p(LPT)C
nmax 3p(OPT)C <
At most two jobs are scheduled on each machine.For such a problem, LPT is optimal.
![Page 21: Uwe Schwiegelshohn EPIT 2007, June 5 Ordonnancementvirtualpanic.com/anonymousftplistings/ebooks/Job Scheduling... · Uwe Schwiegelshohn EPIT 2007, June 5 Ordonnancement. 2 Content](https://reader031.vdocuments.net/reader031/viewer/2022030207/5b9d0ad109d3f2de128b7c6d/html5/thumbnails/21.jpg)
21
A Worst Case Example for LPT
4 parallel machines: P4||Cmax
Cmax(OPT) = 12 =7+5 = 6+6 = 4+4+4Cmax(LPT) = 15 = 11+4=(4/3 -1/(3·4))·12
jobs 1 2 3 4 5 6 7 8 95 5 4 4 46 6pj 7 7
7
7
4
6
4
6
5
5
4
![Page 22: Uwe Schwiegelshohn EPIT 2007, June 5 Ordonnancementvirtualpanic.com/anonymousftplistings/ebooks/Job Scheduling... · Uwe Schwiegelshohn EPIT 2007, June 5 Ordonnancement. 2 Content](https://reader031.vdocuments.net/reader031/viewer/2022030207/5b9d0ad109d3f2de128b7c6d/html5/thumbnails/22.jpg)
22
List Scheduling
LPT requires knowledge of the processing times.No direct application to nonclairvoyant scheduling
Arbitrary nondelay schedule (List Scheduling, Graham, 1966)
Tight approximation factor: m12
(OPT)C(LIST)C
max
max −≤
1 1 1 1 16
1 1 1 1 1
1 1 1 1 11 1 1 1 11 1 1 1 1
1 1 1 1 1
Cmax(LIST)=11
611 1 1 1 111 1 1 1 111 1 1 1 111 1 1 1 111 1 1 1 1
Cmax(OPT)=6
![Page 23: Uwe Schwiegelshohn EPIT 2007, June 5 Ordonnancementvirtualpanic.com/anonymousftplistings/ebooks/Job Scheduling... · Uwe Schwiegelshohn EPIT 2007, June 5 Ordonnancement. 2 Content](https://reader031.vdocuments.net/reader031/viewer/2022030207/5b9d0ad109d3f2de128b7c6d/html5/thumbnails/23.jpg)
23
Online Transformation
Let A be an algorithm for a job scheduling problem without release dates and with
Then there is an algorithm A’ for the corresponding online job scheduling problem with
(Shmoys, Wein, Williamson, 1995)
k(OPT)C
(A)C
max
max ≤
2k(OPT)C
)(A'C
max
max ≤
![Page 24: Uwe Schwiegelshohn EPIT 2007, June 5 Ordonnancementvirtualpanic.com/anonymousftplistings/ebooks/Job Scheduling... · Uwe Schwiegelshohn EPIT 2007, June 5 Ordonnancement. 2 Content](https://reader031.vdocuments.net/reader031/viewer/2022030207/5b9d0ad109d3f2de128b7c6d/html5/thumbnails/24.jpg)
24
Transformation Proof
S0: Jobs available at time 0=F-1=F-2
F0=Cmax(A,S0)Si+1: Jobs released in (Fi-1,Fi]Fi=Cmax(A,Si) such that no job from Si starts before Fi-1.Assume that all jobs in Si are released at time Fi-2
Cmax(OPT) cannot increase while Cmax(A’) remains unchanged.
Proof
)A'(Ck2F maxi ⋅<
)A'(Ck)S A,(CkFFF maximax1-ii2-i ⋅=⋅≤−+
)A'(Ck)S A,(CkFFFFF max1-imax2-i1-i3-i2-i1-i ⋅<⋅≤−+≤−
![Page 25: Uwe Schwiegelshohn EPIT 2007, June 5 Ordonnancementvirtualpanic.com/anonymousftplistings/ebooks/Job Scheduling... · Uwe Schwiegelshohn EPIT 2007, June 5 Ordonnancement. 2 Content](https://reader031.vdocuments.net/reader031/viewer/2022030207/5b9d0ad109d3f2de128b7c6d/html5/thumbnails/25.jpg)
25
List Scheduling Extensions
The List scheduling bound 2-1/m also applies to Pm|rj|Cmax(Hall, Shmoys, 1989).Online extension of List scheduling to parallel jobs:
No machine is kept idle while there is at least one job waiting and there are enough machines idle to start this job (nondelay).
The List scheduling bound 2-1/m also applies to Pm|mj|Cmax (Feldmann, Sgall, Teng, 1994).The List scheduling bound 2-1/m also applies to Pm|mj,rj|Cmax (Naroska, Schwiegelshohn, 2002).
2-1/m is a competitive factor for the corresponding online nonclairvoyant scheduling problem.Proof by induction on the number of different release dates
![Page 26: Uwe Schwiegelshohn EPIT 2007, June 5 Ordonnancementvirtualpanic.com/anonymousftplistings/ebooks/Job Scheduling... · Uwe Schwiegelshohn EPIT 2007, June 5 Ordonnancement. 2 Content](https://reader031.vdocuments.net/reader031/viewer/2022030207/5b9d0ad109d3f2de128b7c6d/html5/thumbnails/26.jpg)
26
Pm | mj | Cmax Proof
The bound holds if during the whole schedule there is no interval with at least m/2 idle machines.
The sum of machines used in any two intervals is larger than m unless the jobs executed in one interval are a subset of the jobs executed in the other interval.
S)(C
m12
1)S(C1-2m
m
S)(C2m
1mpmm1OPT)(C
maxmax
maxjjmax
⋅−
=⋅
≥⋅+
≥⋅≥ ∑
{ }⎭⎬⎫
⎩⎨⎧
⋅⎟⎠⎞
⎜⎝⎛ −⋅⋅⎟
⎠⎞
⎜⎝⎛ −≤ ∑ jjjmax p max
m12 ,pm
m12maxS)(C
![Page 27: Uwe Schwiegelshohn EPIT 2007, June 5 Ordonnancementvirtualpanic.com/anonymousftplistings/ebooks/Job Scheduling... · Uwe Schwiegelshohn EPIT 2007, June 5 Ordonnancement. 2 Content](https://reader031.vdocuments.net/reader031/viewer/2022030207/5b9d0ad109d3f2de128b7c6d/html5/thumbnails/27.jpg)
27
Makespan with Preemptions
Pm |prmp| Cmax is easy.Transformation of a nonpreemptive single machine schedule in a preemptive parallel schedule (McNaughton, 1959)
The single machine schedule is split into at most m schedules oflength Cmax(OPT).Each schedule is executed on a different machine.There are at most m-1 preemptions.
Pm |rj, prmp| Cmax is easy.Longest remaining processing time algorithm.Clairvoyant online scheduling
Competitive factor 1 for allocation as late as possible.Competitive factor e/(e-1)=1.58 for allocation of machine slots at submission time (Chen, van Vliet, Wöginger, 1995)
Nonclairvoyant online scheduling: same competitive factor 2-1/m as for the nonpreemptive case (Shmoys, Wein Williamson, 1995)
![Page 28: Uwe Schwiegelshohn EPIT 2007, June 5 Ordonnancementvirtualpanic.com/anonymousftplistings/ebooks/Job Scheduling... · Uwe Schwiegelshohn EPIT 2007, June 5 Ordonnancement. 2 Content](https://reader031.vdocuments.net/reader031/viewer/2022030207/5b9d0ad109d3f2de128b7c6d/html5/thumbnails/28.jpg)
28
Content of the Lecture
What is job scheduling?
Single machine problems and results
Makespan problems on parallel machines
Utilization problems on parallel machines
Completion time problems on parallel machines
Exemplary workload problem
![Page 29: Uwe Schwiegelshohn EPIT 2007, June 5 Ordonnancementvirtualpanic.com/anonymousftplistings/ebooks/Job Scheduling... · Uwe Schwiegelshohn EPIT 2007, June 5 Ordonnancement. 2 Content](https://reader031.vdocuments.net/reader031/viewer/2022030207/5b9d0ad109d3f2de128b7c6d/html5/thumbnails/29.jpg)
29
Utilization
Utilization Ut is closely related to the makespan Cmax if t=Cmax.
In online job scheduling problems, there is no last submitted job.Ut with t being the actual time is better suited than the makespanobjective.
Pm |rj| UtNonclairvoyant online scheduling: tight competitive factor for any nondelay schedule 1.3333 (Hussein, Schwiegelshohn, 2006)Proof by induction on the different release dates.
2
U2(LIST)=0.75
11
U2(OPT)=1
112
![Page 30: Uwe Schwiegelshohn EPIT 2007, June 5 Ordonnancementvirtualpanic.com/anonymousftplistings/ebooks/Job Scheduling... · Uwe Schwiegelshohn EPIT 2007, June 5 Ordonnancement. 2 Content](https://reader031.vdocuments.net/reader031/viewer/2022030207/5b9d0ad109d3f2de128b7c6d/html5/thumbnails/30.jpg)
30
time
machines
Interval withoutidle machinest2
t1
Utilization Proof (1)
Transformation of the job systemReduction of the release dates
![Page 31: Uwe Schwiegelshohn EPIT 2007, June 5 Ordonnancementvirtualpanic.com/anonymousftplistings/ebooks/Job Scheduling... · Uwe Schwiegelshohn EPIT 2007, June 5 Ordonnancement. 2 Content](https://reader031.vdocuments.net/reader031/viewer/2022030207/5b9d0ad109d3f2de128b7c6d/html5/thumbnails/31.jpg)
31
time
machines
Interval withoutidle machines
Transformation of the job systemSplitting of jobsThe system only contains short and long jobs.
All long jobs start at the end of an interval.
Utilization Proof (2)
![Page 32: Uwe Schwiegelshohn EPIT 2007, June 5 Ordonnancementvirtualpanic.com/anonymousftplistings/ebooks/Job Scheduling... · Uwe Schwiegelshohn EPIT 2007, June 5 Ordonnancement. 2 Content](https://reader031.vdocuments.net/reader031/viewer/2022030207/5b9d0ad109d3f2de128b7c6d/html5/thumbnails/32.jpg)
32
time
machines machinesInterval withoutidle machines
Transformation of the job systemModification of jobs with earlier release dates
Optimal scheduleNondelay schedule
Utilization Proof (3)
![Page 33: Uwe Schwiegelshohn EPIT 2007, June 5 Ordonnancementvirtualpanic.com/anonymousftplistings/ebooks/Job Scheduling... · Uwe Schwiegelshohn EPIT 2007, June 5 Ordonnancement. 2 Content](https://reader031.vdocuments.net/reader031/viewer/2022030207/5b9d0ad109d3f2de128b7c6d/html5/thumbnails/33.jpg)
33
Utilization Proof (4)
If all long jobs of a transformed job system start at their release date, then the utilization is maximal for all t and the equal priority completion time is minimal.
time
machines
short jobs
long jobs
long jobs fromearlier releasedates
![Page 34: Uwe Schwiegelshohn EPIT 2007, June 5 Ordonnancementvirtualpanic.com/anonymousftplistings/ebooks/Job Scheduling... · Uwe Schwiegelshohn EPIT 2007, June 5 Ordonnancement. 2 Content](https://reader031.vdocuments.net/reader031/viewer/2022030207/5b9d0ad109d3f2de128b7c6d/html5/thumbnails/34.jpg)
34
Utilization Proof (5)
r
tσ
Optimal schedule
r
Nondelay schedule S
tστr
τr
tk
ktr
tσ
Optimal schedule
r
Nondelay schedule S
tσ
τk
τr
![Page 35: Uwe Schwiegelshohn EPIT 2007, June 5 Ordonnancementvirtualpanic.com/anonymousftplistings/ebooks/Job Scheduling... · Uwe Schwiegelshohn EPIT 2007, June 5 Ordonnancement. 2 Content](https://reader031.vdocuments.net/reader031/viewer/2022030207/5b9d0ad109d3f2de128b7c6d/html5/thumbnails/35.jpg)
35
Pm | rj,mj | Ut
Parallel jobs may cause intermediate idle time even if all jobs are released at time 0.Nonclairvoyant online scheduling:
Competitive factor → m in the worst case Competitive factor → 2 if the actual time >> max{pj}
2 4 61 3 5
U5(LIST)=0.2+0.16ε
12345
U5(OPT)=1
6
Jobs 1 2 3 4 5 6pj 1+ε 1+ε 1+ε 1+ε 1 5rj 0 1 2 3 4 0mj 1 1 1 1 1 5
![Page 36: Uwe Schwiegelshohn EPIT 2007, June 5 Ordonnancementvirtualpanic.com/anonymousftplistings/ebooks/Job Scheduling... · Uwe Schwiegelshohn EPIT 2007, June 5 Ordonnancement. 2 Content](https://reader031.vdocuments.net/reader031/viewer/2022030207/5b9d0ad109d3f2de128b7c6d/html5/thumbnails/36.jpg)
36
Pm | rj,mj,prmp | Ut
Here, preemption of parallel jobs is based on gang scheduling.
All allocated machines concurrently start, interrupt, resume, and complete the execution of a parallel job.There is no migration or change of parallelism.
Nonclairvoyant online scheduling: competitive factor 4 (Schwiegelshohn, Yahyapour, 2000)
21
3
U3(A)=7/15
14
144
U3(OPT)=14/15
21
4
3
![Page 37: Uwe Schwiegelshohn EPIT 2007, June 5 Ordonnancementvirtualpanic.com/anonymousftplistings/ebooks/Job Scheduling... · Uwe Schwiegelshohn EPIT 2007, June 5 Ordonnancement. 2 Content](https://reader031.vdocuments.net/reader031/viewer/2022030207/5b9d0ad109d3f2de128b7c6d/html5/thumbnails/37.jpg)
37
Content of the Lecture
What is job scheduling?
Single machine problems and results
Makespan problems on parallel machines
Utilization problems on parallel machines
Completion time problems on parallel machines
Exemplary workload problem
![Page 38: Uwe Schwiegelshohn EPIT 2007, June 5 Ordonnancementvirtualpanic.com/anonymousftplistings/ebooks/Job Scheduling... · Uwe Schwiegelshohn EPIT 2007, June 5 Ordonnancement. 2 Content](https://reader031.vdocuments.net/reader031/viewer/2022030207/5b9d0ad109d3f2de128b7c6d/html5/thumbnails/38.jpg)
38
Pm || ΣCj
Pm || ΣCj is easy.Shortest processing time (SPT) (Conway, Maxwell, Miller, 1967)Single machine proof:
Σ Cj=n p(1)+ (n-1) p(2) + … 2 p(n-1) + p(n)p(1) ≤ p(2) ≤ p(3) ≤ ..... ≤ p(n-1) ≤ p(n) must hold for an optimal schedule.
Parallel identical machines proof:Dummy jobs with processing time 0 are added until n is a multiple of m.The sum of the completion time has n additive terms with one coefficient each: m coefficients with value n/m
m coefficients with value n/m – 1 :
m coefficients with value 1If there is one coefficient h>n/m then there must be a coefficient k<n/m. Then we replace h with k+1 and obtain a smaller ΣCj .
Pm |prmp| ΣCj is easy (Shortest remaining processing time).
![Page 39: Uwe Schwiegelshohn EPIT 2007, June 5 Ordonnancementvirtualpanic.com/anonymousftplistings/ebooks/Job Scheduling... · Uwe Schwiegelshohn EPIT 2007, June 5 Ordonnancement. 2 Content](https://reader031.vdocuments.net/reader031/viewer/2022030207/5b9d0ad109d3f2de128b7c6d/html5/thumbnails/39.jpg)
39
Pm || ΣwjCj
Pm || ΣwjCj is strongly NP-hard.The WSPT algorithm has a tight approximation factor of 1.207 (Kawaguchi, Kyan, 1986)It is sufficient to consider instances where all jobs have the same ratio wj/pj.Proof by induction on the number of different ratios.
J is the set of all jobs with the largest ratio in an instance I.The weights of all jobs in J are multiplied by a positive factor ε <1 such that those jobs now have the second largest ratio.This produces instance I’.The WSPT order is still valid.The WSPT schedule remains unchanged.The optimal schedule may change.
![Page 40: Uwe Schwiegelshohn EPIT 2007, June 5 Ordonnancementvirtualpanic.com/anonymousftplistings/ebooks/Job Scheduling... · Uwe Schwiegelshohn EPIT 2007, June 5 Ordonnancement. 2 Content](https://reader031.vdocuments.net/reader031/viewer/2022030207/5b9d0ad109d3f2de128b7c6d/html5/thumbnails/40.jpg)
40
Different Ratios
Induction ProofΣwjCj(WSPT,I’)≤λ・ΣwjCj(OPT,I’) (induction assumption)x: contribution of all jobs in J to ΣwjCj(WSPT,I)y: contribution of all jobs not in J to ΣwjCj(WSPT,I)x’: contribution of all jobs in J to ΣwjCj(OPT,I)y’: contribution of all jobs not in J to ΣwjCj(OPT,I)x≤λ・x’ (induction assumption)ΣwjCj(WSPT,I)= x+y and ΣwjCj(WSPT,I’)=ε・x+y, ΣwjCj(OPT,I)=x’+y’ and ΣwjCj(OPT,I’)≤ε・x’+y’y≤λ・y’ → ΣwjCj(WSPT,I)≤λ・ΣwjCj(OPT,I) y>λ・y’ → λ・x’y>x・λ・y’ → x’/y’>x/y → x’y-xy’>0 → x’y-xy’>ε(x’y-xy’)ΣwjCj(WSPT,I’)・ΣwjCj(OPT,I) =(ε・x+y)(x’+y’)>(ε・x’+y’)(x+y)≥ΣwjCj(OPT,I’)・ΣwjCj(WSPT,I)ΣwjCj(WSPT,I)≤λ・ΣwjCj(OPT,I)
Assumption: wj=pj holds for all jobs j.
![Page 41: Uwe Schwiegelshohn EPIT 2007, June 5 Ordonnancementvirtualpanic.com/anonymousftplistings/ebooks/Job Scheduling... · Uwe Schwiegelshohn EPIT 2007, June 5 Ordonnancement. 2 Content](https://reader031.vdocuments.net/reader031/viewer/2022030207/5b9d0ad109d3f2de128b7c6d/html5/thumbnails/41.jpg)
41
WSPT Proof (1)
time
machines
Transformation of the job systemSplitting of job j into jobs j1 and j2.The system only contains short and long jobs.
All long jobs start at the end of busy interval in the list schedule.
21
22121
2211
jj
jjjjjjjj
jjjjjjjjjj
pp
)S'(Cp)p)S'(C(p)S'(C)pp(
S)(CpS)(Cp)(S'CpS)(Cw)S'(Cw
=
=−−⋅−⋅+=
=−−=−∑∑
∑∑∑∑
∑∑
≥
≥−
−≥
)OPT'(Cw)S'(Cw
pp)OPT'(Cwpp)S'(Cw
OPT)(CwS)(Cw
jj
jj
jjjj
jjjj
jj
jj
21
21
![Page 42: Uwe Schwiegelshohn EPIT 2007, June 5 Ordonnancementvirtualpanic.com/anonymousftplistings/ebooks/Job Scheduling... · Uwe Schwiegelshohn EPIT 2007, June 5 Ordonnancement. 2 Content](https://reader031.vdocuments.net/reader031/viewer/2022030207/5b9d0ad109d3f2de128b7c6d/html5/thumbnails/42.jpg)
42
WSPT Proof (2)
Single machine without intermediate idle timewj=pj holds for all jobs.∑wjCj(S)=∑wjCj(OPT)= 0.5((∑pj)2+∑pj
2)Proof by induction on the number of jobs
( )( ) ( )
( )( ) ( )∑∑∑
∑∑∑∑
++++=
=+++=
2j'
2j
2j'jj'
2j
jj'j'2
j2
jjj
pp21ppp2p
21
ppppp21S)(Cw
![Page 43: Uwe Schwiegelshohn EPIT 2007, June 5 Ordonnancementvirtualpanic.com/anonymousftplistings/ebooks/Job Scheduling... · Uwe Schwiegelshohn EPIT 2007, June 5 Ordonnancement. 2 Content](https://reader031.vdocuments.net/reader031/viewer/2022030207/5b9d0ad109d3f2de128b7c6d/html5/thumbnails/43.jpg)
43
WSPT Proof (3)
Equalization of the long jobsAssumption of a continuous model (fraction of machines)k long jobs with different processing times are transformed inton(k) jobs with the same processing time p(k) such that ∑pj=n(k)・p(k) and ∑pj
2=n(k)・(p(k))2 hold.p(k)= ∑pj
2/ ∑pj and n(k)= (∑pj)2/ ∑pj2
Then we have k≥n(k) for reasons of convexity.
machines
time
machines
![Page 44: Uwe Schwiegelshohn EPIT 2007, June 5 Ordonnancementvirtualpanic.com/anonymousftplistings/ebooks/Job Scheduling... · Uwe Schwiegelshohn EPIT 2007, June 5 Ordonnancement. 2 Content](https://reader031.vdocuments.net/reader031/viewer/2022030207/5b9d0ad109d3f2de128b7c6d/html5/thumbnails/44.jpg)
44
WSPT Proof (4)
machines
time
machines
Modification of the job systemPartitioning of the long jobs into two groupsEqualization of the both groups separatelyThe maximum completion time of the small jobs decreases due to the large rectangle.The jobs of the small rectangle are rearranged.New equalization of the large rectangleDetermination of the size of the large rectangle
![Page 45: Uwe Schwiegelshohn EPIT 2007, June 5 Ordonnancementvirtualpanic.com/anonymousftplistings/ebooks/Job Scheduling... · Uwe Schwiegelshohn EPIT 2007, June 5 Ordonnancement. 2 Content](https://reader031.vdocuments.net/reader031/viewer/2022030207/5b9d0ad109d3f2de128b7c6d/html5/thumbnails/45.jpg)
45
Release Dates
Pm |rj| ΣCj
Approximation factor 2Clairvoyant, randomized online scheduling: competitive factor 2
Pm |rj,prmp| ΣCj
Approximation factor 2Clairvoyant, randomized online scheduling: competitive factor 2
Pm |rj| ΣwjCj
Approximation factor 2Clairvoyant, randomized online scheduling: competitive factor 2
Pm |rj,prmp| ΣwjCj
Approximation factor 2Clairvoyant, randomized online scheduling: competitive factor 2 (all results Schulz, Skutella, 2002)
![Page 46: Uwe Schwiegelshohn EPIT 2007, June 5 Ordonnancementvirtualpanic.com/anonymousftplistings/ebooks/Job Scheduling... · Uwe Schwiegelshohn EPIT 2007, June 5 Ordonnancement. 2 Content](https://reader031.vdocuments.net/reader031/viewer/2022030207/5b9d0ad109d3f2de128b7c6d/html5/thumbnails/46.jpg)
46
Parallel Jobs
Pm |mj,prmp| ΣwjCjUse of gang scheduling without any task migrationApproximation factor 2.37 (Schwiegelshohn, 2004)
Pm |mj,prmp| ΣCjNonclairvoyant approximation factor 2-2/(n+1) if all jobs are malleable with linear speedup (Deng, Gu, Brecht, Lu, 2000).
Pm |mj| ΣwjCjApproximation factor 7.11 (Schwiegelshohn, 2004)Approximation factor 2 if mj≤0.5m holds for all jobs (Turek et al., 1994)
Pm |mj| ΣCjApproximation factor 2 if the jobs are malleable without superlinear speedup (Turek et al., 1994)
![Page 47: Uwe Schwiegelshohn EPIT 2007, June 5 Ordonnancementvirtualpanic.com/anonymousftplistings/ebooks/Job Scheduling... · Uwe Schwiegelshohn EPIT 2007, June 5 Ordonnancement. 2 Content](https://reader031.vdocuments.net/reader031/viewer/2022030207/5b9d0ad109d3f2de128b7c6d/html5/thumbnails/47.jpg)
47
Online Problems
Pm |mj,rj,prmp| ΣwjCj
Nonclairvoyant online scheduling with gang scheduling and wj=mj・pj: competitive factor 3.562 (Schwiegelshohn, Yahyapour, 2000)
wj=mj・pj guarantees that no job is preferred over another job regardless of its resource consumption as all jobs have the same(extended) Smith ratio.All jobs are started in order of their arrival (FCFS).Any job started after a job j can increase the flow time Cj-rj by at most a factor of 2
Clairvoyant online scheduling with malleable jobs and linear speedup:
Competitive factor 12+ε for a deterministic algorithmCompetitive factor 8.67 for a randomized algorithm (both resultsChakrabarti et al.,1996)
![Page 48: Uwe Schwiegelshohn EPIT 2007, June 5 Ordonnancementvirtualpanic.com/anonymousftplistings/ebooks/Job Scheduling... · Uwe Schwiegelshohn EPIT 2007, June 5 Ordonnancement. 2 Content](https://reader031.vdocuments.net/reader031/viewer/2022030207/5b9d0ad109d3f2de128b7c6d/html5/thumbnails/48.jpg)
48
Content of the Lecture
What is job scheduling?
Single machine problems and results
Makespan problems on parallel machines
Utilization problems on parallel machines
Completion time problems on parallel machines
Exemplary workload problem
![Page 49: Uwe Schwiegelshohn EPIT 2007, June 5 Ordonnancementvirtualpanic.com/anonymousftplistings/ebooks/Job Scheduling... · Uwe Schwiegelshohn EPIT 2007, June 5 Ordonnancement. 2 Content](https://reader031.vdocuments.net/reader031/viewer/2022030207/5b9d0ad109d3f2de128b7c6d/html5/thumbnails/49.jpg)
49
MPP Problem
Machine modelMassively parallel processor (MPP): m parallel identical machines
Job modelMultiple independent usersNonclairvoyant (unknown processing time pj ) with estimatesOnline (submission over time rj )Fixed degree of parallelism mj during the whole processingNo preemption
ObjectiveMachine utilizationAverage weighted response time (AWRT): pj・mj・(Cj-rj )Based on user groups
![Page 50: Uwe Schwiegelshohn EPIT 2007, June 5 Ordonnancementvirtualpanic.com/anonymousftplistings/ebooks/Job Scheduling... · Uwe Schwiegelshohn EPIT 2007, June 5 Ordonnancement. 2 Content](https://reader031.vdocuments.net/reader031/viewer/2022030207/5b9d0ad109d3f2de128b7c6d/html5/thumbnails/50.jpg)
50
Algorithmic Approach
Reordering of the waiting queueParameters of jobs in the waiting queueActual timeScheduling situations: weekdays daytime (8am – 6pm), weekdays nighttime (6pm – 8am), weekends
Selected sorting criteriaSelected objective
Consideration of 2 user groups: 10 AWRT1+ 4 AWRT2
Parameter training with Evolution StrategiesRecorded workloads and simulationsWorkload scaling for comparison
Waiting queue
![Page 51: Uwe Schwiegelshohn EPIT 2007, June 5 Ordonnancementvirtualpanic.com/anonymousftplistings/ebooks/Job Scheduling... · Uwe Schwiegelshohn EPIT 2007, June 5 Ordonnancement. 2 Content](https://reader031.vdocuments.net/reader031/viewer/2022030207/5b9d0ad109d3f2de128b7c6d/html5/thumbnails/51.jpg)
51
Workloads and User Groups
Identifier CTC KTH LANL SDSC 00 SDSC 95 SDSC 96
Machine SP2 SP2 CM-5 SP2 SP2 SP2
Period 06/26/96 –05/31/97
09/23/96 –08/29/97
04/10/94 –09/24/96
04/28/98 –04/30/00
12/29/94 –12/30/95
12/27/95 –12/31/96
Processors (m) 1024 1024 1024 1024 1024 1024
Jobs (n) 136471 167375 201378 310745 131762 66185
Workload scaling
User Group 1 2 3 4 5
RCu/RC > 8% 2 – 8 % 1 – 2 % 0.1 – 1 % < 0.1 %
User group definition
![Page 52: Uwe Schwiegelshohn EPIT 2007, June 5 Ordonnancementvirtualpanic.com/anonymousftplistings/ebooks/Job Scheduling... · Uwe Schwiegelshohn EPIT 2007, June 5 Ordonnancement. 2 Content](https://reader031.vdocuments.net/reader031/viewer/2022030207/5b9d0ad109d3f2de128b7c6d/html5/thumbnails/52.jpg)
52
∑=
⎟⎟⎠
⎞⎜⎜⎝
⎛⋅+⋅+⋅=
||
11 )(
Groups
iii processors
imerequestedTbimerequestedT
waitTimeaKwJobf
∑=
⎟⎟⎠
⎞⎜⎜⎝
⎛⋅+⋅+⋅=
||
12 )(
Groups
iii processors
imerequestedTbwaitTimeaKwJobf
( )∑=
⋅⋅+⋅+⋅=||
14 )(
Groups
iii processorsimerequestedTbwaitTimeaKwJobf
∑=
⎟⎟⎠
⎞⎜⎜⎝
⎛⋅
⋅+⋅=||
13 )(
Groups
iii processorsimerequestedT
waitTimeaKwJobf
Sorting Criteria
Training of parameters wi, Ki, a, b with Evolution Strategies
![Page 53: Uwe Schwiegelshohn EPIT 2007, June 5 Ordonnancementvirtualpanic.com/anonymousftplistings/ebooks/Job Scheduling... · Uwe Schwiegelshohn EPIT 2007, June 5 Ordonnancement. 2 Content](https://reader031.vdocuments.net/reader031/viewer/2022030207/5b9d0ad109d3f2de128b7c6d/html5/thumbnails/53.jpg)
53
CTC Training and CTC Workload
-20,00
-15,00
-10,00
-5,00
0,00
5,00
10,00
15,00A
WR
T Im
prov
emen
ts in
%
AWRT 1 AWRT 2
AWRT 3 AWRT 4 AWRT 5
Method AWRT 1 AWRT 2 AWRT 3 AWRT 4 AWRT 5 UTIL
GREEDY 52755.80 s 61947.65 s 56275.18 s 54017.23 s 35085.84 s 66.99 %
EASY 59681.28 s 64976.07 s 50317.47 s 46120.02 s 31855.68 s 66.99 %
![Page 54: Uwe Schwiegelshohn EPIT 2007, June 5 Ordonnancementvirtualpanic.com/anonymousftplistings/ebooks/Job Scheduling... · Uwe Schwiegelshohn EPIT 2007, June 5 Ordonnancement. 2 Content](https://reader031.vdocuments.net/reader031/viewer/2022030207/5b9d0ad109d3f2de128b7c6d/html5/thumbnails/54.jpg)
54
CTC Training and All Workloads
-50,00
-40,00
-30,00
-20,00
-10,00
0,00
10,00
20,00O
bjec
tive
Impr
ovem
ents
in %
CTC
KTH
LANL
SDSC 00
SDSC 95
SDSC 96
Some workloads are similar (CTC, LANL).Some workloads are significantly different (CTC, KTH).
![Page 55: Uwe Schwiegelshohn EPIT 2007, June 5 Ordonnancementvirtualpanic.com/anonymousftplistings/ebooks/Job Scheduling... · Uwe Schwiegelshohn EPIT 2007, June 5 Ordonnancement. 2 Content](https://reader031.vdocuments.net/reader031/viewer/2022030207/5b9d0ad109d3f2de128b7c6d/html5/thumbnails/55.jpg)
55
Results in CTC Paretofront
50000
52000
54000
56000
58000
60000
62000
64000
66000
68000
70000
45000 47000 49000 51000 53000 55000 57000 59000 61000 63000 65000
AWRT 1 in Seconds
AWR
T 2
in S
econ
ds
PFEASYGREEDY CTC optGREEDY ALL opt
![Page 56: Uwe Schwiegelshohn EPIT 2007, June 5 Ordonnancementvirtualpanic.com/anonymousftplistings/ebooks/Job Scheduling... · Uwe Schwiegelshohn EPIT 2007, June 5 Ordonnancement. 2 Content](https://reader031.vdocuments.net/reader031/viewer/2022030207/5b9d0ad109d3f2de128b7c6d/html5/thumbnails/56.jpg)
56
Results in SDSC 95 Paretofront
40000
45000
50000
55000
60000
65000
40000 42000 44000 46000 48000 50000 52000 54000 56000
AWRT 1 in Seconds
AWRT
2 in
Sec
onds
PFEASYGREEDY CTC optGREEDY ALL opt
![Page 57: Uwe Schwiegelshohn EPIT 2007, June 5 Ordonnancementvirtualpanic.com/anonymousftplistings/ebooks/Job Scheduling... · Uwe Schwiegelshohn EPIT 2007, June 5 Ordonnancement. 2 Content](https://reader031.vdocuments.net/reader031/viewer/2022030207/5b9d0ad109d3f2de128b7c6d/html5/thumbnails/57.jpg)
57
Conclusion
Most deterministic job scheduling problems are NP hard.Approximation algorithms
Polynomial time approximation schemes Simple algorithms
Complete problem knowledge is rare in practice.Online algorithms
Competitive analysisStochastic scheduling
Randomized algorithms
ChallengesPartial information
Recorded workloadsUser estimates
Scheduling objectives and constraints