Citation
Resource-Aware Scheduling for Data Centers with Heterogenous Servers. In proceedings of the 7th Multidisciplinary International Conference on Scheduling : Theory and Applications (MISTA 2015), 25 - 28 Aug 2015, Prague, Czech Republic, pages 240-259, 2015.
Paper
Abstract
This paper presents an algorithm for resource-aware scheduling of computational jobs in a large-scale heterogeneous data center. The algorithm aims to allocate di?erent machine con?gurations to job classes to attain an e?cient mapping between job resource request pro?les and machine resource capacity pro?les. We propose a three-stage algorithm. The ?rst stage uses a queueing model that treats the system in an aggregated manner with pooled machines and jobs represented as a ?uid ?ow. The latter two stages use combinatorial optimization techniques to take the solution from the ?rst stage and apply it to a more accurate representation of the data center. In the second stage, jobs and machines are discretized. A linear programming model is created to obtain a solution to the discrete problem that maximizes the system capacity. The third and ?nal stage is a scheduling policy that uses the solution from the second stage to guide the dispatching of arriving jobs to machines. Using Google workload trace data, we show that our algorithm outperforms a benchmark greedy dispatch policy. We ?nd that our algorithm is able to provide mean response times up to an order of magnitude smaller than the benchmark dispatch policy. These results show that it is important to consider the heterogeneity of machine con?guration pro?les in making e?ective scheduling decisions.
You can download the pdf of this publication from here
doi
This publication does not have a doi, so we cannot provide a link to the original source
What is a doi?: A doi (Document Object Identifier) is a unique identifier for sicientific papers (and occasionally other material). This provides direct access to the location where the original article is published using the URL http://dx.doi/org/xxxx (replacing xxx with the doi). See http://dx.doi.org/ for more information
URL
This pubication does not have a URL associated with it.
The URL is only provided if there is additional information that might be useful. For example, where the entry is a book chapter, the URL might link to the book itself.
Bibtex
@INPROCEEDINGS{2015-240-259-P, author = {T. T. Tran and P. Y. Zhang and H. Li and D. G. Down and J.C. Beck},
title = {Resource-Aware Scheduling for Data Centers with Heterogenous Servers},
booktitle = {In proceedings of the 7th Multidisciplinary International Conference on Scheduling : Theory and Applications (MISTA 2015), 25 - 28 Aug 2015, Prague, Czech Republic},
year = {2015},
editor = {Z. Hanzalek and G. Kendall and B. McCollum and P. Sucha},
pages = {240--259},
note = {Paper},
abstract = { This paper presents an algorithm for resource-aware scheduling of computational jobs in a large-scale heterogeneous data center. The algorithm aims to allocate di?erent machine con?gurations to job classes to attain an e?cient mapping between job resource request pro?les and machine resource capacity pro?les. We propose a three-stage algorithm. The ?rst stage uses a queueing model that treats the system in an aggregated manner with pooled machines and jobs represented as a ?uid ?ow. The latter two stages use combinatorial optimization techniques to take the solution from the ?rst stage and apply it to a more accurate representation of the data center. In the second stage, jobs and machines are discretized. A linear programming model is created to obtain a solution to the discrete problem that maximizes the system capacity. The third and ?nal stage is a scheduling policy that uses the solution from the second stage to guide the dispatching of arriving jobs to machines. Using Google workload trace data, we show that our algorithm outperforms a benchmark greedy dispatch policy. We ?nd that our algorithm is able to provide mean response times up to an order of magnitude smaller than the benchmark dispatch policy. These results show that it is important to consider the heterogeneity of machine con?guration pro?les in making e?ective scheduling decisions.},
owner = {Graham},
timestamp = {2017.01.16},
webpdf = {2015-240-259-P.pdf} }