Scheduling with Testing: Competitive Algorithms for Minimizing the Total Weighted Completion Time in the Adversarial Model
Pith reviewed 2026-06-25 21:46 UTC · model grok-4.3
The pith
Constant-competitive algorithms exist for scheduling with testing to minimize total weighted completion time on single and parallel machines.
A machine-rendered reading of the paper's core claim, the machinery that carries it, and where it could break.
Core claim
The authors establish the first constant-competitive algorithms for weighted scheduling with testing in the adversarial model. For single-machine scheduling they present a deterministic algorithm with competitive ratio 2.3166 and a randomized variant with ratio 2.1523; these match the best-known upper bounds for the unweighted setting. Combining the same single-machine procedures with list scheduling yields competitive ratios of 2.7763 (deterministic) and 2.5110 (randomized) for identical parallel machines, improving even the unweighted case.
What carries the argument
The central mechanism is a threshold-based decision rule (with randomization in one variant) that decides whether to test each job or run it to its upper bound, followed by list scheduling on the revealed processing times; the analysis bounds the weighted completion time against an optimal offline schedule by comparing the cost incurred on tested versus untested jobs.
If this is right
- The competitive ratios remain unchanged when job weights are introduced, matching the unweighted guarantees.
- Parallel-machine bounds follow directly from applying the single-machine procedures inside a list-scheduling wrapper.
- The same algorithmic framework improves the best-known ratios even when all weights are set to one.
- The results hold in the fully adversarial online model where processing times are revealed only upon testing.
Where Pith is reading between the lines
- The testing decision thresholds could be tuned for other machine environments or objectives such as makespan.
- If testing times themselves are uncertain, the current analysis would need adjustment but the overall competitive approach might still apply.
- The matching of weighted and unweighted ratios suggests that the proof technique may extend to other linear cost functions on completion times.
Load-bearing premise
Testing a job always reveals its exact processing time with no noise, and the algorithm must commit online to testing or running without knowledge of future jobs or the adversary's choices.
What would settle it
An explicit job instance with weights, upper bounds, and testing times on which the algorithm's total weighted completion time exceeds 2.3166 times the optimal offline cost (or the stated randomized or parallel ratios) would falsify the claimed competitive ratios.
read the original abstract
We study scheduling with testing on a single machine and on identical parallel machines to minimize the total \emph{weighted} completion time in the adversarial model. In this setting, each job is equipped with a weight, an upper bound on its processing time, and a testing time. An algorithm can either execute a job for an amount of time equal to the upper bound or test it first to reveal a potentially lower processing time used to schedule the job later. We establish the first constant-competitive algorithms for this problem with job-dependent weights that reflect each job's relative importance. For single-machine scheduling, we present a deterministic algorithm with a competitive ratio of 2.3166 and show that a randomized variant has a competitive ratio of 2.1523. These guarantees match the best-known upper bounds in the unweighted setting. Combining these algorithms with list scheduling yields competitive ratios of 2.7763 and 2.5110 for identical-parallel-machine scheduling, improving the previously best-known bounds even in the unweighted case.
Editorial analysis
A structured set of objections, weighed in public.
Referee Report
Summary. The manuscript studies scheduling with testing on a single machine and identical parallel machines to minimize total weighted completion time in the adversarial model, where each job has a weight, an upper bound on processing time, and a testing time. An online algorithm may either run a job to its upper bound or test it to reveal its exact (lower) processing time. The paper claims the first constant-competitive algorithms for the weighted setting: a deterministic single-machine algorithm with ratio 2.3166, a randomized variant with ratio 2.1523 (matching the best unweighted bounds), and parallel-machine ratios of 2.7763 (deterministic) and 2.5110 (randomized) obtained by combining the single-machine algorithms with list scheduling, which also improves the prior unweighted parallel-machine bounds.
Significance. If the stated competitive ratios are established by the analysis, the work is significant because it is the first to obtain constant competitiveness for job-dependent weights (rather than unit weights) while matching the best-known unweighted ratios; the parallel-machine improvement via a standard list-scheduling reduction is a clean and useful extension. The adversarial model with exact revelation on testing is the standard one in the area, and the explicit numerical ratios allow direct comparison with prior work.
minor comments (3)
- [§3] §3 (single-machine deterministic algorithm): the description of how the weight-dependent ordering is maintained after testing should be expanded with a short pseudocode fragment or invariant statement, as the current prose leaves the tie-breaking rule for equal weights ambiguous.
- [§5] §5 (parallel-machine analysis): the application of list scheduling to the single-machine schedules is stated at a high level; a one-paragraph derivation showing that the 4/3 factor from list scheduling multiplies cleanly with the single-machine ratios (without additional cross terms) would improve readability.
- [Table 1] Table 1 (competitive-ratio summary): the table lists the new ratios but omits the previous best unweighted parallel-machine bounds being improved; adding those numbers would make the improvement claim immediately verifiable.
Simulated Author's Rebuttal
We thank the referee for the positive summary, significance assessment, and recommendation of minor revision. No specific major comments appear in the report, so we have nothing further to address point by point.
Circularity Check
No significant circularity identified
full rationale
The paper presents competitive-ratio proofs for online scheduling algorithms against an optimal offline solution (OPT), which is an external benchmark independent of the algorithm's own outputs or fitted parameters. The claimed ratios (2.3166 deterministic, 2.1523 randomized) are derived via standard list-scheduling and randomization techniques that compare algorithm cost directly to OPT cost; no step reduces a prediction to a fitted input, renames a known result, or relies on a self-citation chain for the core bound. The extension from unweighted to weighted cases is presented as carrying through the same analysis structure without redefining quantities in terms of themselves. This is the normal, non-circular structure for competitive-analysis papers.
Axiom & Free-Parameter Ledger
Reference graph
Works this paper leans on
-
[1]
Albers and A
S. Albers and A. Eckl. Explorable uncertainty in scheduling with non-uniform testing times. In C. Kaklamanis and A. Levin, editors,WAOA 2020, volume 12806 ofLNCS, pages 127–142, Cham,
2020
-
[2]
Springer.doi:10.1007/978-3-030-80879-2_9
-
[3]
Albers and A
S. Albers and A. Eckl. Scheduling with testing on multiple identical parallel machines. In A. Lubiw, M. Salavatipour, and M. He, editors,WADS 2021, volume 12808 ofLNCS, pages 29–42, Cham,
2021
-
[4]
Springer.doi:10.1007/978-3-030-83508-8_3
-
[5]
A. Amouzandeh, K. Jansen, L. Pirotton, R. van Stee, and C. Wambsganz. Minimizing the weighted makespan with restarts on a single machine, 2025.doi:10.48550/arXiv.2510.09589
-
[6]
Kononov, Giorgio Lucarelli, and Fanny Pascual
E. Bampis, K. Dogeas, A. Kononov, G. Lucarelli, and F. Pascual. Speed scaling with explorable uncertainty. InProceedings of the 33rd ACM Symposium on Parallelism in Algorithms and Architec- tures (SPAA 2021), pages 83–93, New York, NY, USA, 2021. Association for Computing Machinery. doi:10.1145/3409964.3461812
-
[7]
Brucker.Scheduling Algorithms
P. Brucker.Scheduling Algorithms. Springer, Berlin, Heidelberg, 5 edition, 2007.doi:10.1007/ 978-3-540-69516-5
2007
-
[8]
F. Buld and A. S. Schulz. Scheduling with testing: Competitive algorithms for minimizing the total weighted completion time in the adversarial model. In V. Chau, C. D¨ urr, M. Li, and P. Lu, editors,Frontiers of Algorithmics, volume 15828 ofLNCS, pages 64–77, Singapore, 2025. Springer. doi:10.1007/978-981-96-8312-3_5
-
[9]
Damerius, P
C. Damerius, P. Kling, M. Li, C. Xu, and R. Zhang. Scheduling with a limited testing budget: Tight results for the offline and oblivious settings. In I. L. Gørtz, M. Farach-Colton, S. J. Puglisi, and G. Herman, editors,31st Annual European Symposium on Algorithms (ESA 2023), volume 274 ofLeibniz International Proceedings in Informatics (LIPIcs), pages 38:...
2023
-
[10]
Schloss Dagstuhl – Leibniz-Zentrum f¨ ur Informatik.doi:10.4230/LIPIcs.ESA.2023.38
-
[11]
Scheduling with obligatory tests
K. Dogeas, T. Erlebach, and Y.-C. Liang. Scheduling with obligatory tests. In T. Chan, J. Fischer, J. Iacono, and G. Herman, editors,32nd Annual European Symposium on Algo- rithms (ESA 2024), volume 308 ofLeibniz International Proceedings in Informatics (LIPIcs), pages 48:1–48:14, Dagstuhl, Germany, 2024. Schloss Dagstuhl – Leibniz-Zentrum f¨ ur Informati...
-
[12]
F. Dufoss´ e, C. D¨ urr, N. Nadal, D. Trystram, and´O. C. V´ asquez. Scheduling with a processing time oracle.Applied Mathematical Modelling, 104:701–720, 2022.doi:10.1016/j.apm.2021.12.020
-
[13]
An adversarial model for scheduling with testing
C. D¨ urr, T. Erlebach, N. Megow, and J. Meißner. An adversarial model for scheduling with testing. Algorithmica, 82(12):3630–3675, 2020.doi:10.1007/s00453-020-00742-2
-
[14]
W. L. Eastman, S. Even, and I. M. Isaacs. Bounds for the optimal scheduling ofnjobs onm processors.Management Science, 11(2):268–279, 1964.doi:10.1287/mnsc.11.2.268
-
[15]
M. Gong, Z.-Z. Chen, G. Chen, G. Lin, and L. Wang. Randomized algorithms for fully online multiprocessor scheduling with testing.Annals of Operations Research, 358(1):71–101, 2026.doi: 10.1007/s10479-025-06804-4
-
[16]
M. Gong, Z.-Z. Chen, and K. Hayashi. Approximation algorithms for multiprocessor scheduling with testing to minimize the total job completion time.Algorithmica, 86(5):1400–1427, 2024.doi: 10.1007/s00453-023-01198-w
-
[17]
M. Gong, J. Fan, G. Lin, B. Su, Z. Su, and X. Zhang. Multiprocessor scheduling with testing: improved online algorithms and numerical experiments.Journal of Scheduling, 28(5):513–527, 2025. doi:10.1007/s10951-025-00850-3
-
[18]
M. Gong, R. Goebel, G. Lin, and E. Miyano. Improved approximation algorithms for non-preemptive multiprocessor scheduling with testing.Journal of Combinatorial Optimization, 44(1):877–893, 2022. doi:10.1007/s10878-022-00865-y
-
[19]
J.-H. Kim and K.-Y. Chwa. Non-clairvoyant scheduling for weighted flow time.Information Pro- cessing Letters, 87(1):31–37, 2003.doi:10.1016/S0020-0190(03)00231-X. 15
-
[20]
R. Levi, T. Magnanti, and Y. Shaposhnik. Scheduling with testing.Management Science, 65(2):776– 793, 2019.doi:10.1287/mnsc.2017.2973
-
[21]
R. Levi, T. Magnanti, and Y. Shaposhnik. Scheduling with testing of heterogeneous jobs.Manage- ment Science, 70(5):2934–2953, 2024.doi:10.1287/mnsc.2023.4833
-
[22]
The Completion-Threshold Framework for Obligatory-Test Scheduling on Multiple Machines
K.-C. Liang and Y.-C. Liang. The completion-threshold framework for obligatory-test scheduling on multiple machines, 2026.doi:10.48550/arXiv.2606.02029
work page internal anchor Pith review Pith/arXiv arXiv doi:10.48550/arxiv.2606.02029 2026
-
[23]
A. Lindermayr, G. Sch¨ afer, J. Schl¨ oter, and L. Stougie. Online flow time minimization with gradually revealed jobs, 2026.doi:10.48550/arXiv.2602.12716
-
[24]
A. H.-H. Liu, F.-H. Liu, P. W. Wong, and X.-O. Zhang. The power of amortization on scheduling with explorable uncertainty. In J. Byrka and A. Wiese, editors,WAOA 2023, volume 14297 of LNCS, pages 90–103, Cham, 2023. Springer.doi:10.1007/978-3-031-49815-2_7
-
[25]
A. H.-H. Liu, F.-H. Liu, P. W. Wong, and X.-O. Zhang. Amortization helps for scheduling with explorable uncertainty, even on parallel machines. InProceedings of the 16th Workshop on Models and Algorithms for Planning and Scheduling Problems (MAPSP 2024), pages 259–262, 2024
2024
-
[26]
R. H. M¨ ohring, A. S. Schulz, and M. Uetz. Approximation in stochastic scheduling: The power of lp-based priority policies.Journal of the ACM, 46(6):924–942, 1999.doi:10.1145/331524.331530
-
[27]
R. Motwani, S. Phillips, and E. Torng. Nonclairvoyant scheduling.Theoretical Computer Science, 130(1):17–47, 1994.doi:10.1016/0304-3975(94)90151-1
-
[28]
M. L. Pinedo.Scheduling: Theory, Algorithms, and Systems. Springer, Cham, 6 edition, 2022. doi:10.1007/978-3-031-05921-6
-
[29]
W. E. Smith. Various optimizers for single-stage production.Naval Research Logistics Quarterly, 3(1-2):59–66, 1956.doi:10.1002/nav.3800030106
-
[30]
Z. Sun, N. T. Argon, and S. Ziya. Patient triage and prioritization under austere conditions. Management Science, 64(10):4471–4489, 2018.doi:10.1287/mnsc.2017.2855
-
[31]
X. Zhang. Scheduling with explorable uncertainty for minimising the total weighted completion time. Master’s thesis, Utrecht University, 2023. A Unboundedness of Previous Approaches Considering Job- Dependent Weights For scheduling with testing to minimize the totalweightedcompletion time in the adversarial model, some algorithms were proposed by [28]. Th...
2023
-
[32]
Then applyDelay-All individually to each batch in order of non-increasing weights
Intermediate strategyL-Delay-All: Group jobs in batches of same weight. Then applyDelay-All individually to each batch in order of non-increasing weights. They provide upper bounds on the competitive ratios of these algorithms, parameterized by input values such as the upper bounduor the maximal weightw max. It also contains lower bounds of 2.4 for the se...
-
[33]
, n, (long)t j = 1, uj =n 2, pj =n 2, wj = 1 +εforj=n+ 1
AlgorithmsGreedyandL-Delay-All: Consider instances withn≥2smalljobs, onelargejob and ε >0 small: (short)t j = 1, uj =n 2, pj = 0, w j = 1 forj= 1, . . . , n, (long)t j = 1, uj =n 2, pj =n 2, wj = 1 +εforj=n+ 1. 16 Since the first algorithm focuses only on the weight, it processes the long job immediately after testing, causing a large delay for all short ...
-
[34]
, n, (heavy)t j = 1, uj = 2, pj = 0, wj =n 2 forj=n+ 1
AlgorithmDelay-All: Consider instances withnlightjobs and oneheavyjob: (light)t j = 1, uj = 2, pj = 0, wj = 1 forj= 1, . . . , n, (heavy)t j = 1, uj = 2, pj = 0, wj =n 2 forj=n+ 1. Delay-Alltests all jobs first and then executes them. Especially the heavy job is not completed immediately after testing. The optimal offline algorithm tests and executes each...
-
[35]
Remark B.3.TheWeighted(α, β)-SORTalgorithm can be extended to identical parallel machines as theWeighted(α, β)-PCPalgorithm in Section 4
To improve upon this, one would needβ < √ 2 since we have the term 1 +βin the maximum, henceα < √ 2 since we have the termα(1 + 1/β), but then 1 + 2/α>1 + √ 2. Remark B.3.TheWeighted(α, β)-SORTalgorithm can be extended to identical parallel machines as theWeighted(α, β)-PCPalgorithm in Section 4. Using the pairwise contribution bound from Theorem B.2 and ...
discussion (0)
Sign in with ORCID, Apple, or X to comment. Anyone can read and Pith papers without signing in.