Using Run-Time Predictions to Estimate Queue Wait Times and Improve Scheduler Performance

Abstract

On many computers, a request to run a job is not serviced immediately but instead is placed in a queue and serviced only when resources are r eleased b y p r eceding jobs. In this paper, we build on run-time prediction techniques that we developed i n p r evious research to explore two problems. The rst problem is to predict how long applications will wait in a queue until they receive resources. We show that run-time estimates can be used for this and that using our run-time estimates result in more a c curate wait-time predictions than when the run-time prediction techniques of other researches are used. The second problem we investigate is improving scheduling performance. We use run-time predictions to improve the performance of the least work rst and back ll scheduling algorithms. We nd that using our run-time predictor results in lower mean wait times for the workloads with higher o ered l o ads when compared to alternative run-time predictors.

Measuring the Performance and Reliability of Production Computational Grids

“…We can see in the tables that the two leading causes of variability (i.e., the highest standard deviations relative to the means) are: (i) initialization and finalization operations on the "login" node (initialization, staging data, building executable, and cleanup); and (ii) queue wait times. This was expected as the number of users actively working on a login node varies greatly over time and queue wait times are known to exhibit variable and non-stationary behaviors [9]. As seen in Table IV we show three separate totals: (i) total probe runtime; (ii) probe runtime excluding setup/cleanup; and (iii) probe runtime excluding setup/cleanup and queue wait times.…”

An effort prediction framework for software defect correction

“…After experimentation, the authors conclude that the Top-K Mean method performed better than any of the three regression approaches (linear, logarithmic and inverse). Similarly, in [75], Smith et al use Top-K Linear Regression to predict run times for applications on computational grid systems. Yet again, their results suggest that Top-K Mean performs better than th e Top-K Linear Regression approach.…”

A Grid Resource Broker Supporting Advance Reservations and Benchmark-Based Resource Selection

“…The performance differences between Grid resources and the fact that their relative performance characteristics may vary for different types of applications makes resource selection difficult, see, e.g., [9,16,19,20]. Our approach to handle this is to use a benchmark-based procedure for resource selection.…”