A comparative study of real workload traces and synthetic workload models for non-contiguous allocation in 2D meshes

Saad Bani Mohammad, Ismail Ababneh, Ould Khaoua Mohamed

Research output: Chapter in Book/Report/Conference proceedingConference contribution

Abstract

The performance of the existing non-contiguous processor allocation strategies has been traditionally carried out by means of simulation based on synthetic workload models to generate streams of incoming jobs. To validate the performance of the existing algorithms, there is need to evaluate the algorithms' performance based on real workload traces. In this paper, we evaluate the performance of several well-known processor allocation strategies based on real workload traces and compare the results against those obtained using synthetic workload models. Our results reveal that the Greedy Available Busy List allocation strategy (GABL) performs better than all other non-contiguous allocation strategies in terms of job turnaround time for synthetic workload models, whereas the performance of Multiple Buddy Strategy (MBS) for real workload traces is superior to that of the other allocation strategies. This is because MBS performs well for power of two job sizes and contiguous allocation is explicitly sought in MBS for requests with sizes of the power of two. Moreover, the results indicate that the relative performance merits of the non-contiguous GABL strategy over the remaining non-contiguous allocation strategies become more noticeable as the message length increases.

Original languageEnglish
Title of host publicationInternational Conference on Scalable Computing and Communications - The 8th International Conference on Embedded Computing, ScalCom-EmbeddedCom 2009
Pages633-639
Number of pages7
DOIs
Publication statusPublished - 2009
EventInternational Conference on Scalable Computing and Communications- 8th International Conference on Embedded Computing, ScalCom-EmbeddedCom 2009 - Dalian, China
Duration: Sep 25 2009Sep 27 2009

Other

OtherInternational Conference on Scalable Computing and Communications- 8th International Conference on Embedded Computing, ScalCom-EmbeddedCom 2009
CountryChina
CityDalian
Period9/25/099/27/09

Fingerprint

Turnaround time

ASJC Scopus subject areas

  • Computational Theory and Mathematics
  • Computer Science Applications
  • Software

Cite this

Mohammad, S. B., Ababneh, I., & Mohamed, O. K. (2009). A comparative study of real workload traces and synthetic workload models for non-contiguous allocation in 2D meshes. In International Conference on Scalable Computing and Communications - The 8th International Conference on Embedded Computing, ScalCom-EmbeddedCom 2009 (pp. 633-639). [5341885] https://doi.org/10.1109/EmbeddedCom-ScalCom.2009.121

A comparative study of real workload traces and synthetic workload models for non-contiguous allocation in 2D meshes. / Mohammad, Saad Bani; Ababneh, Ismail; Mohamed, Ould Khaoua.

International Conference on Scalable Computing and Communications - The 8th International Conference on Embedded Computing, ScalCom-EmbeddedCom 2009. 2009. p. 633-639 5341885.

Research output: Chapter in Book/Report/Conference proceedingConference contribution

Mohammad, SB, Ababneh, I & Mohamed, OK 2009, A comparative study of real workload traces and synthetic workload models for non-contiguous allocation in 2D meshes. in International Conference on Scalable Computing and Communications - The 8th International Conference on Embedded Computing, ScalCom-EmbeddedCom 2009., 5341885, pp. 633-639, International Conference on Scalable Computing and Communications- 8th International Conference on Embedded Computing, ScalCom-EmbeddedCom 2009, Dalian, China, 9/25/09. https://doi.org/10.1109/EmbeddedCom-ScalCom.2009.121
Mohammad SB, Ababneh I, Mohamed OK. A comparative study of real workload traces and synthetic workload models for non-contiguous allocation in 2D meshes. In International Conference on Scalable Computing and Communications - The 8th International Conference on Embedded Computing, ScalCom-EmbeddedCom 2009. 2009. p. 633-639. 5341885 https://doi.org/10.1109/EmbeddedCom-ScalCom.2009.121
Mohammad, Saad Bani ; Ababneh, Ismail ; Mohamed, Ould Khaoua. / A comparative study of real workload traces and synthetic workload models for non-contiguous allocation in 2D meshes. International Conference on Scalable Computing and Communications - The 8th International Conference on Embedded Computing, ScalCom-EmbeddedCom 2009. 2009. pp. 633-639
@inproceedings{0f4425496cb64c0fa69f31a2025401c2,
title = "A comparative study of real workload traces and synthetic workload models for non-contiguous allocation in 2D meshes",
abstract = "The performance of the existing non-contiguous processor allocation strategies has been traditionally carried out by means of simulation based on synthetic workload models to generate streams of incoming jobs. To validate the performance of the existing algorithms, there is need to evaluate the algorithms' performance based on real workload traces. In this paper, we evaluate the performance of several well-known processor allocation strategies based on real workload traces and compare the results against those obtained using synthetic workload models. Our results reveal that the Greedy Available Busy List allocation strategy (GABL) performs better than all other non-contiguous allocation strategies in terms of job turnaround time for synthetic workload models, whereas the performance of Multiple Buddy Strategy (MBS) for real workload traces is superior to that of the other allocation strategies. This is because MBS performs well for power of two job sizes and contiguous allocation is explicitly sought in MBS for requests with sizes of the power of two. Moreover, the results indicate that the relative performance merits of the non-contiguous GABL strategy over the remaining non-contiguous allocation strategies become more noticeable as the message length increases.",
author = "Mohammad, {Saad Bani} and Ismail Ababneh and Mohamed, {Ould Khaoua}",
year = "2009",
doi = "10.1109/EmbeddedCom-ScalCom.2009.121",
language = "English",
isbn = "9780769538259",
pages = "633--639",
booktitle = "International Conference on Scalable Computing and Communications - The 8th International Conference on Embedded Computing, ScalCom-EmbeddedCom 2009",

}

TY - GEN

T1 - A comparative study of real workload traces and synthetic workload models for non-contiguous allocation in 2D meshes

AU - Mohammad, Saad Bani

AU - Ababneh, Ismail

AU - Mohamed, Ould Khaoua

PY - 2009

Y1 - 2009

N2 - The performance of the existing non-contiguous processor allocation strategies has been traditionally carried out by means of simulation based on synthetic workload models to generate streams of incoming jobs. To validate the performance of the existing algorithms, there is need to evaluate the algorithms' performance based on real workload traces. In this paper, we evaluate the performance of several well-known processor allocation strategies based on real workload traces and compare the results against those obtained using synthetic workload models. Our results reveal that the Greedy Available Busy List allocation strategy (GABL) performs better than all other non-contiguous allocation strategies in terms of job turnaround time for synthetic workload models, whereas the performance of Multiple Buddy Strategy (MBS) for real workload traces is superior to that of the other allocation strategies. This is because MBS performs well for power of two job sizes and contiguous allocation is explicitly sought in MBS for requests with sizes of the power of two. Moreover, the results indicate that the relative performance merits of the non-contiguous GABL strategy over the remaining non-contiguous allocation strategies become more noticeable as the message length increases.

AB - The performance of the existing non-contiguous processor allocation strategies has been traditionally carried out by means of simulation based on synthetic workload models to generate streams of incoming jobs. To validate the performance of the existing algorithms, there is need to evaluate the algorithms' performance based on real workload traces. In this paper, we evaluate the performance of several well-known processor allocation strategies based on real workload traces and compare the results against those obtained using synthetic workload models. Our results reveal that the Greedy Available Busy List allocation strategy (GABL) performs better than all other non-contiguous allocation strategies in terms of job turnaround time for synthetic workload models, whereas the performance of Multiple Buddy Strategy (MBS) for real workload traces is superior to that of the other allocation strategies. This is because MBS performs well for power of two job sizes and contiguous allocation is explicitly sought in MBS for requests with sizes of the power of two. Moreover, the results indicate that the relative performance merits of the non-contiguous GABL strategy over the remaining non-contiguous allocation strategies become more noticeable as the message length increases.

UR - http://www.scopus.com/inward/record.url?scp=72449200621&partnerID=8YFLogxK

UR - http://www.scopus.com/inward/citedby.url?scp=72449200621&partnerID=8YFLogxK

U2 - 10.1109/EmbeddedCom-ScalCom.2009.121

DO - 10.1109/EmbeddedCom-ScalCom.2009.121

M3 - Conference contribution

SN - 9780769538259

SP - 633

EP - 639

BT - International Conference on Scalable Computing and Communications - The 8th International Conference on Embedded Computing, ScalCom-EmbeddedCom 2009

ER -