From the second half of the seventies and throughout the eighties, queuing network models, based upon the so-called class of “BCMP networks”, have been successfully applied to various real environments to support development efforts in the early stage of design for both computer-communication systems and flexible manufacturing systems. A significant boost to the practical application came from the celebrated “MVA algorithm” for the analytical solution of (closed) queuing networks, once that the initial difficulties (numerical stability) under load-dependent servers were solved. However, whenever the queuing network at hand was refined to match with detailed realistic features of the domain do be modelled, it was (and still is) necessary to switch to the resolution by means of discrete-event simulation, in order to provide reliable confidence intervals upon the expected values of the performance metrics of interest. With reference to the application domains above mentioned, the major model features detected since then as very difficult to handle were: (1) service interruptions and server vacations; (2) FIFO services under class-dependent service times, even exponentially distributed; (3) non pre-emptive priority service disciplines for customers belonging to different classes; (4) fork and join mechanisms for circulating customers; (5) blocking phenomena different from rejection blocking with random rerouting; (6) state-dependent routing of circulating customers; (7) passive resources and population constrained subnetworks. To manage these difficulties, approximate solutions based on variants of the original MVA algorithm for BCMP networks were proposed, applied and evaluated against discrete-event simulation. Special attention was paid to the Norton theorem for queuing networks and to the flow-equivalent reduction of a given subnetwork with a state-dependent exponential server. Finally, infinite server stations, acting as a source for finite population closed models, were often replaced by external Poissonian flows of customers entering the network: despite the lack of the celebrated “arrival theorem” upon which the original MVA algorithm is founded, Poissonian arrivals were found sometimes to guarantee approximately the same average number of circulating customers as in the original closed network.Later on, at the end of the nineties, several Windows-based environments for easy development, verification and validation of simulators, aimed at reproducing in great detail the queuing-based reality of interest with a limited amount of programming effort, became available to practitioners. Hence, the easy to implement and read characteristic of both exact and approximate MVA based analytical formulas and algorithms became less appealing as opposite to the friendly usage of commercial simulators. So, the question is: why bother today? Why should research efforts towards new analytical queuing approximations, as stimulated by old and new real-world domains, be encouraged not only for academic purposes?Nowadays, the (relatively) new domain of maritime container terminals asks for renewed research efforts to achieve approximate, yet fast analytical solutions of queuing networks tailored upon this exciting domain, which is rich of further non-standard queuing features difficult to be addressed. A set of specific queuing systems are presented in this talk with respect to a set of queuing processes and starvation/blocking phenomena detected within the execution of the main logistic processes at a maritime container terminal devoted to pure transshipment. Non-standard modelling features asking for new approximate analytical solution methods are highlighted and discussed with a special focus on (i) vessel arrival and priorities in port admittance and berthing, (ii) container discharge/loading processes by means of rail-mounted gantry cranes, (iii) container transfer back and forth from the yard to the quay by means of human-operated vehicles and (iv) container storage in and retrieval from several yard blocks where set-down and pick-up operations within a large number of parallel storage rows call for a special care when modelling the arrival-access-departure process of vehicles (straddle carriers) affected by mutual exclusion phenomena and queuing before service or re-routing. We focus on the round-trip process executed by a fleet of straddle carriers at the pure transshipment port of reference that involves container discharge/loading from/on vessels and transfer from/to berth bollards and yard rows; containers are represented as passive resources to be gathered and delivered by a finite population of straddle carriers. A special purpose discrete-event simulation model has been designed and implemented to analyze the accuracy of some obliged choices adopted for the (approximate) analytical solution of the queuing network models for the above round-trip process: the first choice being the classical Poisson process despite of the bulk arrival processes of containers; the second being the adoption of a pure delay station to represent the travel of human-operated straddle carriers from any given bollard along the quay to any given row within the yard and vice versa; the third being the forced waiting status of any given straddle carrier who finds the destination yard row already occupied by another straddle carrier. More generally, we focus on the effectiveness of some MVA based (analytical) approximations to the closed queuing network at hand, under both low and high variance properties of service activities carried out by straddle carriers on berth and on travel from the berth to the yard and vice versa. Starvation and blocking phenomena experienced by straddle carriers, as well as queuing phenomena during the travel along the internal routes of a real terminal of reference are also evaluated by numerical experiments with our simulation model. This model is used to first support the development of some queuing approximations as first and then validate the usage of a closed queuing network versus an alternative open setting.
Queuing network models in port logistics
LEGATO Pasquale;MAZZA Rina Mary
2014-01-01
Abstract
From the second half of the seventies and throughout the eighties, queuing network models, based upon the so-called class of “BCMP networks”, have been successfully applied to various real environments to support development efforts in the early stage of design for both computer-communication systems and flexible manufacturing systems. A significant boost to the practical application came from the celebrated “MVA algorithm” for the analytical solution of (closed) queuing networks, once that the initial difficulties (numerical stability) under load-dependent servers were solved. However, whenever the queuing network at hand was refined to match with detailed realistic features of the domain do be modelled, it was (and still is) necessary to switch to the resolution by means of discrete-event simulation, in order to provide reliable confidence intervals upon the expected values of the performance metrics of interest. With reference to the application domains above mentioned, the major model features detected since then as very difficult to handle were: (1) service interruptions and server vacations; (2) FIFO services under class-dependent service times, even exponentially distributed; (3) non pre-emptive priority service disciplines for customers belonging to different classes; (4) fork and join mechanisms for circulating customers; (5) blocking phenomena different from rejection blocking with random rerouting; (6) state-dependent routing of circulating customers; (7) passive resources and population constrained subnetworks. To manage these difficulties, approximate solutions based on variants of the original MVA algorithm for BCMP networks were proposed, applied and evaluated against discrete-event simulation. Special attention was paid to the Norton theorem for queuing networks and to the flow-equivalent reduction of a given subnetwork with a state-dependent exponential server. Finally, infinite server stations, acting as a source for finite population closed models, were often replaced by external Poissonian flows of customers entering the network: despite the lack of the celebrated “arrival theorem” upon which the original MVA algorithm is founded, Poissonian arrivals were found sometimes to guarantee approximately the same average number of circulating customers as in the original closed network.Later on, at the end of the nineties, several Windows-based environments for easy development, verification and validation of simulators, aimed at reproducing in great detail the queuing-based reality of interest with a limited amount of programming effort, became available to practitioners. Hence, the easy to implement and read characteristic of both exact and approximate MVA based analytical formulas and algorithms became less appealing as opposite to the friendly usage of commercial simulators. So, the question is: why bother today? Why should research efforts towards new analytical queuing approximations, as stimulated by old and new real-world domains, be encouraged not only for academic purposes?Nowadays, the (relatively) new domain of maritime container terminals asks for renewed research efforts to achieve approximate, yet fast analytical solutions of queuing networks tailored upon this exciting domain, which is rich of further non-standard queuing features difficult to be addressed. A set of specific queuing systems are presented in this talk with respect to a set of queuing processes and starvation/blocking phenomena detected within the execution of the main logistic processes at a maritime container terminal devoted to pure transshipment. Non-standard modelling features asking for new approximate analytical solution methods are highlighted and discussed with a special focus on (i) vessel arrival and priorities in port admittance and berthing, (ii) container discharge/loading processes by means of rail-mounted gantry cranes, (iii) container transfer back and forth from the yard to the quay by means of human-operated vehicles and (iv) container storage in and retrieval from several yard blocks where set-down and pick-up operations within a large number of parallel storage rows call for a special care when modelling the arrival-access-departure process of vehicles (straddle carriers) affected by mutual exclusion phenomena and queuing before service or re-routing. We focus on the round-trip process executed by a fleet of straddle carriers at the pure transshipment port of reference that involves container discharge/loading from/on vessels and transfer from/to berth bollards and yard rows; containers are represented as passive resources to be gathered and delivered by a finite population of straddle carriers. A special purpose discrete-event simulation model has been designed and implemented to analyze the accuracy of some obliged choices adopted for the (approximate) analytical solution of the queuing network models for the above round-trip process: the first choice being the classical Poisson process despite of the bulk arrival processes of containers; the second being the adoption of a pure delay station to represent the travel of human-operated straddle carriers from any given bollard along the quay to any given row within the yard and vice versa; the third being the forced waiting status of any given straddle carrier who finds the destination yard row already occupied by another straddle carrier. More generally, we focus on the effectiveness of some MVA based (analytical) approximations to the closed queuing network at hand, under both low and high variance properties of service activities carried out by straddle carriers on berth and on travel from the berth to the yard and vice versa. Starvation and blocking phenomena experienced by straddle carriers, as well as queuing phenomena during the travel along the internal routes of a real terminal of reference are also evaluated by numerical experiments with our simulation model. This model is used to first support the development of some queuing approximations as first and then validate the usage of a closed queuing network versus an alternative open setting.I documenti in IRIS sono protetti da copyright e tutti i diritti sono riservati, salvo diversa indicazione.
