Non-clairvoyant Reduction Algorithms for Heterogeneous Platforms
نویسندگان
چکیده
We revisit the classical problem of the reduction collective operation in a heterogeneous environment. We discuss and evaluate four algorithms that are non-clairvoyant, i.e., they do not know in advance the computation and communication costs. On the one hand, Binomial-stat and Fibonacci-stat are static algorithms that decide in advance which operations will be reduced, without adapting to the environment; they were originally defined for homogeneous settings. On the other hand, Tree-dyn and Non-Commut-Tree-dyn are fully dynamic algorithms, for commutative or non-commutative reductions. With identical computation costs, we show that these algorithms are approximation algorithms. When costs are exponentially distributed, we perform an analysis of Tree-dyn based on Markov chains. Finally, we assess the relative performance of all four non-clairvoyant algorithms with heterogeneous costs though a set of simulations. Key-words: scheduling, reduction, approximation algorithms, non-clairvoyant algorithms ∗ École Normale Supérieure de Lyon and IUF, Lyon, France † FEMTO-ST, Université de Franche-Comté, Besançon, France ‡ CNRS, École Normale Supérieure de Lyon, Lyon, France ha l-0 08 32 10 2, v er si on 3 11 J un 2 01 3 Algorithmes de réductions non-clairvoyants pour plates-formes hétérogènes Résumé : Nous revisitons le problème classique de la primitive de communication collective de réduction dans le cadre d’un environnement hétérogène. Nous présentons et analysons quatre algorithmes non-clairvoyants, c’est-à-dire qui ne connaissent pas les coûts de calcul et de communication. D’un côté, Binomialstat et Fibonacci-stat sont des algorithmes statiques qui décident a priori de l’arbre de réduction, sans pouvoir s’adapter à l’environnement au cours de la réduction. D’un autre côté, Tree-dyn et Non-Commut-Tree-dyn sont des algorithmes complètement dynamiques pour les réductions commutatives ou non-commutatives. Lorsque les coûts de calcul sont constants, nous montrons que les algorithmes commutatifs admettent des facteurs d’approximation. Lorsque les coûts de calcul et/ou de communication sont distribués selon une loi exponentielle, nous analysons Tree-dyn à l’aide des chaînes de Markov. Enfin, nous comparons les performances des quatre algorithmes dans un environnement hétérogène à l’aide de simulations. Mots-clés : ordonnancement, réduction, algorithmes d’approximation, algorithmes non-clairvoyants ha l-0 08 32 10 2, v er si on 3 11 J un 2 01 3 Non-clairvoyant reduction algorithms for heterogeneous platforms 3
منابع مشابه
Sleep with Guilt and Work Faster to Minimize Flow Plus Energy
In this paper we extend the study of flow-energy scheduling to a model that allows both sleep management and speed scaling. Our main result is a sleep management algorithm called IdleLonger, which works online for a processor with one or multiple levels of sleep states. The design of IdleLonger is interesting; among others, it may force the processor to idle or even sleep even though new jobs h...
متن کاملScheduling heterogeneous processors isn't as easy as you think
We consider preemptive online scheduling algorithms to minimize the total weighted/unweighted flow time plus energy for speed-scalable heterogeneous multiprocessors. We show that the well-known priority scheduling algorithms Highest Density First, Weighted Shortest Elapsed Time First, and Weighted Late Arrival Processor Sharing, are not O(1)-speed O(1)-competitive for the objective of weighted ...
متن کاملWorkflow Fairness Control on Online and Non-clairvoyant Distributed Computing Platforms
Fairly allocating distributed computing resources among workflow executions is critical to multi-user platforms. However, this problem remains mostly studied in clairvoyant and offline conditions, where task durations on resources are known, or the workload and available resources do not vary along time. We consider a non-clairvoyant, online fairness problem where the platform workload, task co...
متن کاملSemi-clairvoyant Scheduling
In [2] it was shown that the obvious semi-clairvoyant generalization of the Shortest Processing Time is O(1)-competitive with respect to average stretch on a single machine. In [2] it was left as an open question whether it was possible for a semiclairvoyant algorithm to be O(1)-competitive with respect to average flow time on a single machine. Here we settle this open question by giving a semi...
متن کاملNon-Clairvoyant Multiprocessor Scheduling of Jobs with Changing Execution Characteristics
A multiprocessor system is unlikely to have access to information about the execution characteristics of the jobs it is to schedule. In this work, we are interested in scheduling algorithms for batch jobs that require no such knowledge (such algorithms are called non-clairvoyant). Preemptive scheduling (i.e., redistribution of processors) is important to reduce mean response time in multiproces...
متن کامل