Copy
Hisao KAMEDA, "Performance Requirements Realizable by Scheduling Strategies in Time-Shared Systems" in IEICE TRANSACTIONS on transactions,
vol. E69-E, no. 12, pp. 1334-1341, December 1986, doi: .
Abstract: Realization of performance requirements by scheduling strategies is investigated on a finite-source queueing model, which has often been used as a model of time-shared systems. The model consists of a single-server station (processor) and a multiple-server station (terminals); each job has a distinct mean service time at the processor. Performance requirements that are stated in terms of the performance vector, each element of which is the utilization factor of the processor for a job, are investigated, and the following results are obtained. First, the optimization of performance measures is discussed under Markovian assumptions. Linear performance measures such as the throughput and the mean response time can be optimized by preemptive priority disciplines, and their implementation may be simple. However, the optimization of nonlinear performance measures such as the average cost per interaction may lead to a linearly constrained nonlinear programming problem, and the schedules which optimize them may be hard to implement. Second, realization of the equal response ratio is discussed. The processor sharing (or preemptive-resume LCFS) discipline, although it attains the equal response ratio in M/G/1, gives a smaller response ratio to a job with a longer mean processor time. It is shown under Markovian assumptions that the equal response ratio is achievable by another scheduling strategy.
URL: https://global.ieice.org/en_transactions/transactions/10.1587/e69-e_12_1334/_p
Copy
@ARTICLE{e69-e_12_1334,
author={Hisao KAMEDA, },
journal={IEICE TRANSACTIONS on transactions},
title={Performance Requirements Realizable by Scheduling Strategies in Time-Shared Systems},
year={1986},
volume={E69-E},
number={12},
pages={1334-1341},
abstract={Realization of performance requirements by scheduling strategies is investigated on a finite-source queueing model, which has often been used as a model of time-shared systems. The model consists of a single-server station (processor) and a multiple-server station (terminals); each job has a distinct mean service time at the processor. Performance requirements that are stated in terms of the performance vector, each element of which is the utilization factor of the processor for a job, are investigated, and the following results are obtained. First, the optimization of performance measures is discussed under Markovian assumptions. Linear performance measures such as the throughput and the mean response time can be optimized by preemptive priority disciplines, and their implementation may be simple. However, the optimization of nonlinear performance measures such as the average cost per interaction may lead to a linearly constrained nonlinear programming problem, and the schedules which optimize them may be hard to implement. Second, realization of the equal response ratio is discussed. The processor sharing (or preemptive-resume LCFS) discipline, although it attains the equal response ratio in M/G/1, gives a smaller response ratio to a job with a longer mean processor time. It is shown under Markovian assumptions that the equal response ratio is achievable by another scheduling strategy.},
keywords={},
doi={},
ISSN={},
month={December},}
Copy
TY - JOUR
TI - Performance Requirements Realizable by Scheduling Strategies in Time-Shared Systems
T2 - IEICE TRANSACTIONS on transactions
SP - 1334
EP - 1341
AU - Hisao KAMEDA
PY - 1986
DO -
JO - IEICE TRANSACTIONS on transactions
SN -
VL - E69-E
IS - 12
JA - IEICE TRANSACTIONS on transactions
Y1 - December 1986
AB - Realization of performance requirements by scheduling strategies is investigated on a finite-source queueing model, which has often been used as a model of time-shared systems. The model consists of a single-server station (processor) and a multiple-server station (terminals); each job has a distinct mean service time at the processor. Performance requirements that are stated in terms of the performance vector, each element of which is the utilization factor of the processor for a job, are investigated, and the following results are obtained. First, the optimization of performance measures is discussed under Markovian assumptions. Linear performance measures such as the throughput and the mean response time can be optimized by preemptive priority disciplines, and their implementation may be simple. However, the optimization of nonlinear performance measures such as the average cost per interaction may lead to a linearly constrained nonlinear programming problem, and the schedules which optimize them may be hard to implement. Second, realization of the equal response ratio is discussed. The processor sharing (or preemptive-resume LCFS) discipline, although it attains the equal response ratio in M/G/1, gives a smaller response ratio to a job with a longer mean processor time. It is shown under Markovian assumptions that the equal response ratio is achievable by another scheduling strategy.
ER -