Uniform Bounds for Scheduling with Job Size Estimates

@inproceedings{Scully2021UniformBF,
  title={Uniform Bounds for Scheduling with Job Size Estimates},
  author={Ziv Scully and Isaac Grosof and Michael Mitzenmacher},
  booktitle={Information Technology Convergence and Services},
  year={2021}
}
We consider the problem of scheduling to minimize mean response time in M/G/1 queues where only estimated job sizes (processing times) are known to the scheduler, where a job of true size s has estimated size in the interval [ βs, αs ] for some α ≥ β > 0. We evaluate each scheduling policy by its approximation ratio , which we define to be the ratio between its mean response time and that of Shortest Remaining Processing Time (SRPT), the optimal policy when true sizes are known. Our question: is… 

Figures from this paper

SEH: Size Estimate Hedging Scheduling of Queues

For a single server system, Shortest Remaining Processing Time (SRPT) is an optimal size-based policy. In this paper, we discuss scheduling a single-server system when exact information about the

Incentive Compatible Queues Without Money

For job scheduling systems, where jobs require some amount of processing and then leave the system, it is natural for each user to provide an estimate of their job’s time requirement in order to aid

Balancing SRPT and FCFS via Starvation Mitigation

The algorithm signicantly improves upon SRPT and FCFS in terms of the competitive ratio for minimizing the ℓ 2 norm of ow time, even if the estimate is loose.

Balanced Allocations with the Choice of Noise

This work proves that for any $m \geq n$ the gap is $O(g+\log n)$ with high probability, and generalizes the heavily loaded analysis of [BCSV06, TW14] for the Two-Choice process, and establishes that asymptotically the same gap bound holds even if many (or possibly all) load comparisons among"similarly loaded"bins are wrong.

Permutation Predictions for Non-Clairvoyant Scheduling

This work proposes a new prediction model, which provides a relative order of jobs which could be seen as predicting algorithmic actions rather than parts of the unknown input, and presents the first learning-augmented scheduling results for weighted jobs and unrelated machines.

Non-Clairvoyant Scheduling with Predictions Revisited

This work proposes a new prediction model, which provides a relative order of jobs which could be seen as predicting algorithmic actions rather than parts of the unknown input, and presents the first learning-augmented scheduling results for weighted jobs and unrelated machines.

Learning Predictions for Algorithms with Predictions

A general design approach is introduced that identifies a functional dependence of the performance measure on the prediction quality and applies techniques from online learning to learn predictors, tune robustness-consistency trade-offs, and bound the sample complexity.

A Universal Error Measure for Input Predictions Applied to Online Graph Problems

A novel measure for quantifying the error in input predictions is introduced based on a minimum-cost hyperedge cover in a suitably dened hypergraph and provides a general template which is applied to online graph problems.

Bankrupting DoS Attackers Despite Uncertainty

An EDoS defense, LCharge, is proposed, that uses resource burning while accounting for on-demand provisioning and retains a strong security guarantee despite the uncertainty of this estimate of the number of job requests from honest clients.

Private Algorithms with Private Predictions

This work proposes to use the algorithms with predictions framework as a powerful way of designing and analyzing privacy-preserving methods that can take advantage of external information to improve utility and demonstrates how to enable differentially private algorithms to make use of and learn noisy predictions.

References

SHOWING 1-10 OF 28 REFERENCES

Scheduling despite inexact job-size information

This work derive upper and lower bounds on the mean response time, the mean slowdown, the response-time tail, and the conditional response time of ε-SMART policies, which provide designers insight into how accurate job-size information must be in order to achieve desired performance guarantees.

PSBS: Practical Size-Based Scheduling

Practical Size-Based Scheduler has near-optimal performance in a large variety of cases with inaccurate size information, that it performs fairly and that it handles job weights correctly and is maintained that it could inspire the design of schedulers in a wide array of real-world use cases.

Nearly insensitive bounds on SMART scheduling

This work proves simple upper and lower bounds on mean response time for all policies in the SMART class and shows that the response times of SMART policies are largely insensitive to the variability of the job size distribution.

Nonclairvoyant scheduling to minimize the total flow time on single and parallel machines

It is proved that a randomized version of the Multilevel Feedback algorithm is competitive for single and parallel machine systems, in the opinion providing one theoretical validation of the goodness of an idea that has proven effective in practice along the last two decades.

The Gittins Policy is Nearly Optimal in the M/G/k under Extremely General Conditions

This work gives the first general analysis of Gittins in the M/G/k and shows that under extremely general conditions, Gittin's mean response time is at most itsmean response time in theM/g/1 plus an $O(łog(1/(1 - ρ)$ additive term, where ρ is the system load.

Minimizing flow time nonclairvoyantly

RMLF is a randomized nonclairvoyant algorithm that has competitive ratio /spl theta/(lognloglogn) against an adaptive adversary and is a slight variation of the multi level feedback (MLF) algorithm used by the Unix operating system, further justifying the adoption of this algorithm.

Performance Modeling and Design of Computer Systems: Queueing Theory in Action

Tackling the questions that systems designers care about, this book brings queueing theory decisively back to computer science and helps readers acquire the skills needed to model, analyze, and design large-scale systems with good performance and low cost.

The Gittins Policy in the M/G/1 Queue

  • Ziv ScullyMor Harchol-Balter
  • Computer Science
    2021 19th International Symposium on Modeling and Optimization in Mobile, Ad hoc, and Wireless Networks (WiOpt)
  • 2021
This work provides the first fully general proof of Gittins’s optimality in the M/G/1, and the optimality result obtained is even more general than was previously known.

Scheduling with Predictions and the Price of Misprediction

This work considers the setting where the service time is not known, but is predicted by for example a machine learning algorithm, and derives formulae for the performance of several strategies for queueing systems that use predictions for service times in order to schedule jobs.

SOAP Bubbles: Robust Scheduling Under Adversarial Noise

This paper studies scheduling for the M/G/1 when the scheduler knows only approximate job ages, finding that naively using the traditional Gittins policy is not robust, meaning that introducing even an infinitesimal amount of noise in job ages can cause a large jump in mean response time.