Upload
others
View
3
Download
0
Embed Size (px)
Citation preview
Greed Is Good(For Scheduling Under Uncertainty)
Marc [email protected]
(updated) IPCO 2017 paper, with:
Varun Gupta, Ben Moseley, Qiaomin Xie
1-Slide Overview
Talk is about (basic, notorious) stochastic scheduling problem[defined later]
Almost all previous results (scheduling in general):
good algorithms based on solving (complicated LP) relaxations
Alternative title: To hell with (LP-)relaxations!
if greedy (online) algorithms are good enoughfirst results (for this model) where jobs appear online
Greedy algorithm has performance guarantee(144 + 72∆)(72 + 36∆)(12 + 6∆)(6 + 3∆)h(∆)
[∆ = (squared) coeff. variation, 1 ≤ h( · ) ≤ 2 with h(0) = 1 ]
Marc Uetz - Greed. . . is good 2
“Greed, . . . , greed is good!”
Marc Uetz - Greed. . . is good 3
Unrelated Machine Scheduling – R | |∑
wjCj
Given: m machines, n non-preemptive jobs with weights wj andmachine-dependent processing times pij :
Cj := completion time job j in schedule; minimize∑
j wj Cj
0 timeCblue
Theorem
Offline: problem is APX-hard [Hoogeveen et al., 2002]
Offline: LP-based 1.4994-approximation [Li, 2018]
Online: lower bound 1.309 [Vestjens 1997]
Marc Uetz - Greed. . . is good 4
Uncertainty in Scheduling
Job j appears at release time rj
Non-clairvoyant online models (job lengths unknown till Cj)
- hopeless, since jobs cannot be preempted[Ω(n) even in simplest settings]
Clairvoyant online models (job lengths known upon arrival rj)
- deterministic algo.: 8-competitive[Hall et al. MOR 1997]
- randomized algo.: 5.78-competitive[Chakrabarti et al. ICALP 1996]
Stochastic Online Model (= Data Science)
Non-clairvoyant, but probabilistic info on pij[our result ⇒ Greedy is deterministic 6-competitive algo.]
Marc Uetz - Greed. . . is good 5
1 Stochastic Scheduling
2 Greedy Algorithm for Unrelated Machines
3 Final Remarks
Marc Uetz - Greed. . . is good 6
Stochastic Processing Times
job j appears: size = (independent) random variables Pij ; known
1
0time t
Pr[Pij ≥ t]
Solution: Non-anticipatory scheduling policy Π
Decisions may only use information up to now.
0 timenow
Objective: Min. expected performance E[∑
wjCj ] =∑
wjE[Cj ]
Marc Uetz - Greed. . . is good 7
Example with Four Jobs
n = 4 jobs, all weights wj = 1
time0 1 10
blue jobs: Pj = 1
green jobs: Pj =
0 probability 4/5
10 probability 1/5(E[Pj ] = 2)
Optimal policy for m = 2 identical machines?
Marc Uetz - Greed. . . is good 8
Policies are Complex (and Dynamic)
Unique optimal policy: Start green + blue
Then continue:
first green then blue if first green job = 0
first blue then green if first green job = 10
[with E[∑
j wjCj ] = 6.76]
0 1 2 11 1210
Complicated tradeoff between
small E[Pj ] or large Pr(Pj = 0) (but, heavy tail)
Marc Uetz - Greed. . . is good 9
Approximation for Stochastic Scheduling
Optimal policies hopeless, even offline, . . .
Definition (Approximation)
Policy Π has performance guarantee α ≥ 1, if for allinstances P
E[∑
wjCΠj ] ≤ α E[
∑wjC
OPTj ]
Adversary OPT knows set of jobs, but subject to uncertainprocessing times Pij , too
Classical competitive analysis is special case
Marc Uetz - Greed. . . is good 10
Coefficient of Variation
Performance guarantees depend on “variability” of Pij
Define
∆ := maxi ,j CV[Pij ]2 = Var[Pij ] / E2[Pij ]
Marc Uetz - Greed. . . is good 11
Approximation Algorithms Stochastic Scheduling
For identical machines:
Mohring, Schulz & U. [J.ACM, 1999]
first LP-based approximation algorithme.g.: Smith’s rule (wj/E[Pj ]) has guarantee ( 3+∆
2 )
improved to 1 + 12 (√
2− 1)(1 + ∆) [Jager & Skutella 2018]
Skutella & U. [SICOMP, 2005], Megow, U. & Vredeveld [Math. OR,
2006] Chou et al. [OR 2006], Schulz [COCOA 2008]
Problems w. precedence constraints, release times, or online
Im, Moseley, Pruhs [STACS, 2015]
remarkable O( log2 n + m log n )-approximation
For unrelated machines:
Skutella, Sviridenko, U. [Math. OR, 2016]
( 3+∆2 ) for offline, using time-indexed LP relaxation
Marc Uetz - Greed. . . is good 12
Greedy for Unrelated Machines, Stochastic Jobs
Theorem
If all release times rj = 0 (“online-list”), Greedy has
performance guarantee 4 + 2∆,
analysis tight for ∆ = 0 [Correa & Queyranne, 2012].
Theorem
In general (release times rj > 0, “online-time”), Greedy has
performance guarantee (6 + 3∆)h(∆) .
1 ≤ h( · ) ≤ 2 with h(0) = 1, h(1) = 3/2
Marc Uetz - Greed. . . is good 13
Greedy Algorithm I
sequencing: available jobs w. max. wj/E[Pij ] go first
assignment: use “proxy” for E[increase] of objective: job jappears ar rj , consider all jobs that arrived earlier, theirmachine assignments fixed, and assuming pij = E[Pij ] andrk = 0 for all jobs, compute expected increase of
∑j wjE[Cj ]
if j was inserted into sequence in order of wj/E[Pij ]
→ assign job j to any machine minimizing this increase
j
0
i
Marc Uetz - Greed. . . is good 14
Greedy Algorithm II
sequencing: available jobs w. max. wj/E[Pij ] go first
availability: job j declared available on machine i at slightlyinflated release time rij ≥ rj
rij = maxE[Pij ], sj
where sj = start time of job j in nominal Greedy schedulewhere we assume pik = E[Pik ] for all jobs on i[note: this can be computed online]
rj rij
j
0
i
Marc Uetz - Greed. . . is good 15
Sketch Analysis
1. time-indexed LP Relaxation (for stochastic problem) - sorry
2. “simplify” that LP relaxation – losing O( ∆ )
3. analysis of Greedy using dual LP solution – losing O( 1 )
“dual fitting” [Anand et al., SODA 2012]
Marc Uetz - Greed. . . is good 16
1 - Time-Indexed LP
yijt := Pr[machine i has job j in process at [t, t + 1)]
0 1 2 11 1210
second, green job (say j = 3), has at time t = 2
y1,3,2 = 4/25y2,3,2 = 1/25
Marc Uetz - Greed. . . is good 17
1 - LP Relaxation (Not a Formulation!)
zS := min∑j∈J
wj CSj
s.t. CSj =
∑i∈M
∑t≥0
(yijt
E[Pij ]
(t + 1
2
)+
1− CV[Pij ]21− CV[Pij]
2
2yijt
)∑i∈M
∑t≥rj
yijtE[Pij ]
= 1 ∀ jobs j ,
∑j∈J
yijt ≤ 1 ∀ machines i , times t,
∑i∈M
∑t≥rj
yijt ≤ CSj ∀ jobs j ,
yijt ≥ 0 ∀ jobs j , machines i , times t.
Would like to work with (LP) dual, but. . .
Marc Uetz - Greed. . . is good 18
2 - Simplified LP Relaxation
zP := min∑j∈J
wj CPj
s.t. CPj =
∑i∈M
∑t≥0
(yijt
E[Pij ]
(t + 1
2
)+
1
2yijt
)∑i∈M
∑t≥0
yijtE[Pij ]
= 1 jobs j ,
∑j∈J
yijt ≤ 1 machines i , times t,
yijt ≥ 0 jobs j , machines i , times t.
Lemma
zP ≤(1 +
∆
2
)zS
Marc Uetz - Greed. . . is good 19
2 - LP Dual
the dual has variables (α, β);
max zD =∑j∈J
αj −∑i∈M
∑t≥0
βit
s.t. αj ≤ E[Pij ]βit + wj
(t +
E[Pij ]
2+
1
2
)for all i , j , t
βit ≥ 0 for all i , t
Lemma
Considering Greedydet (for pij = E[pij ]); can construct feasibledual solution (α, β) with
zD(α, β) =1
6Greedydet
[idea: 3x speed augmentation]Marc Uetz - Greed. . . is good 20
3 - Final Steps
By duality
Greedydet = 6zD(α, β) ≤ 6zD=6zP
≤ 6(1 +∆
2)zS ≤ (6 + 3∆)OPT
Finally, can show for (true) expected starting time job j
Lemma
E [Sj ] ≤ h(∆)︸ ︷︷ ︸≤2
sdetj
sdetj = starting time job j in Greedydet
Proof:
Sj ≤ sj +∑
predecessors k
(Pk − E[Pk ])+ . . .
Marc Uetz - Greed. . . is good 21
Final Remarks
O( ∆ ) is tight (for greedy), there is a ∆/2 lower bound
open problems
1. is const. approximation (indep. of ∆) possible?
2. is stochastic problem harder to approximate ?
thanks for your attention!
Marc Uetz - Greed. . . is good 22