An extremal property of the fifo discipline via an ordinal version of
- 1 January 1989
- journal article
- research article
- Published by Informa UK Limited in Communications in Statistics. Stochastic Models
- Vol. 5 (3), 515-529
- https://doi.org/10.1080/15326348908807121
Abstract
We apply the relation to prove that in great generality the long-run average number of customers in a queueing system at an arrival epoch is equal to the long-run average number of arrivals during the period a customer spends in the system. This relation can be regarded as an ordinal version of , arising when we measure time solely in terms of the number of arrivals that occur. We apply the ordinal version of to obtain a conservation law for single-server queues. We use this conservation law to establish an extremal property for the FIFO service discipline: We show that in a G/GI/I system the FIFO discipline minimizes (maximizes) the long-run average sojourn time per customer among all work-conserving disciplines that are non-anticipating with respect to the service times (may depend on completed service times, but not remaining service times) when the service-time distribution is NBUE (NWUE). Among the disciplines in this class are round robin, processor sharing and shortest expected remaining processing timeKeywords
This publication has 12 references indexed in Scilit:
- M/G/∞ with batch arrivalsOperations Research Letters, 1988
- Response times in M/M/1 time-sharing schemes with limited number of service positionsJournal of Applied Probability, 1988
- Server sharing with a limited number of service positions and symmetric queuesJournal of Applied Probability, 1987
- An Optimal Design Problem for Limited Processor Sharing SystemsManagement Science, 1987
- A central-limit-theorem version ofL=?wQueueing Systems, 1986
- Deciding Which Queue to Join: Some CounterexamplesOperations Research, 1986
- The sojourn time in the GI/M/1 queue by processor sharingJournal of Applied Probability, 1984
- Poisson Arrivals See Time AveragesOperations Research, 1982
- Technical Note—A Last Word on L = λWOperations Research, 1974
- A Proof for the Queuing Formula: L = λWOperations Research, 1961