Fast Pareto Front Approximation for Cloud Instance Pool Optimization

Computing the Pareto Set (PS) of optimal cloud schedules in terms of cost and makespan for a given application and set of cloud instance types is NP-complete. Moreover, cloud instances' volatility requires fast PS recomputations. While genetic algorithms (GA) are a promising approach, little knowledge of an approximated PS's quality leads to GAs running for overly many generations, contradicting the goal of quickly computing an approximate solution. We address this with MOO-GA, our GA enhanced with a domain-tailored termination criteria delivering fast, well-approximated Pareto sets. We compare to NSGAIII using PS convergence and diversity, and computational effort metrics. Results show MOO-GA consistently computing better quality Pareto sets within one second on average (df=98, p-value<10-3).

[1]  Jesús García,et al.  A stopping criterion based on Kalman estimation techniques with several progress indicators , 2009, GECCO.

[2]  Heike Trautmann,et al.  A Convergence Criterion for Multiobjective Evolutionary Algorithms Based on Systematic Statistical Testing , 2008, PPSN.

[3]  Lothar Thiele,et al.  Quality Assessment of Pareto Set Approximations , 2008, Multiobjective Optimization.

[4]  Thilo Kielmann,et al.  Budget Estimation and Control for Bag-of-Tasks Scheduling in Clouds , 2011, Parallel Process. Lett..

[5]  Thilo Kielmann,et al.  Fast (re-)configuration of mixed on-demand and spot instance pools for high-throughput computing , 2013, ORMaCloud '13.

[6]  Kalyanmoy Deb,et al.  An Evolutionary Many-Objective Optimization Algorithm Using Reference-Point-Based Nondominated Sorting Approach, Part I: Solving Problems With Box Constraints , 2014, IEEE Transactions on Evolutionary Computation.

[7]  Hiroyuki Sato,et al.  Inverted PBI in MOEA/D and its impact on the search performance on multi and many-objective optimization , 2014, GECCO.

[8]  Jessica Andrea Carballido,et al.  On Stopping Criteria for Genetic Algorithms , 2004, SBIA.

[9]  Lothar Thiele,et al.  Multiobjective genetic programming: reducing bloat using SPEA2 , 2001, Proceedings of the 2001 Congress on Evolutionary Computation (IEEE Cat. No.01TH8546).

[10]  Gary B. Lamont,et al.  Multiobjective evolutionary algorithms: classifications, analyses, and new innovations , 1999 .

[11]  Shapour Azarm,et al.  Metrics for Quality Assessment of a Multiobjective Design Optimization Solution Set , 2001 .