On Reservation Systems and Queueing Systems Mark E. McGinley, Xiangfei Zhu, and Malathi Veeraraghavan University of Virginia Charlottesville, VA 22904-4743 Abstract—Systems in which resources are shared can be op- erated either with reservation schedulers or as queueing systems. While reservation systems are the only option to support future start-time requests, both types of systems can be used when users want resources at the earliest available opportunity. The objective of this work is to determine the conditions under which a reservation scheduler is required versus when a queueing system is sufficient. The problem was motivated by a communication network service. A new scheduled dynamic circuit service that uses a reservation scheduler is being offered. It differs from plain old telephony service (POTS), which is operated as a queueing sys- tem. Three models of reservation systems are analyzed: (i) finite- window reservation system model, (ii) infinite-window reservation system model, and (iii) synchronized-server model. Our main finding is that if the number of servers is small, a reservation scheduler is required in both blocking-mode and waiting-mode systems. In blocking-mode systems, users should specify multiple start-time options in reservation requests to lower blocking rate while simultaneously operating the system at high utilization. In waiting-mode systems with small numbers of servers, since high-utilization system operation will cause large waiting times, a reservation system is preferred as it will allow customers to carry out other activities between their times of request and allocated start times rather than wait in a queue. I. Introduction The purpose of this work is to understand the differences between reservation systems and queueing systems, and to determine the conditions under which a reservation system should be used. Our motivation for this study is the application of these two types of systems to communication networks. Specifically, we consider circuit and virtual circuit (VC) net- works, which require resource allocation before user data exchange. After providing background on the use of queueing and reservation systems for resource allocation in circuit/VC networks, we formulate our problem statement, and describe our solution approach. Key findings are summarized next. Background and related work. Plain-old telephony service (POTS) is offered on a circuit-switched network. A circuit is established using a call setup procedure, which is initiated when a user dials a telephone number. The telephone number, carried in a call-setup signaling message, is used by the call processing (software) engine within each circuit switch to determine the next-hop switch toward which to the route the call in order to reach the destination. Each call processing engine determines if a DS0 (64 kbps) channel is available on a link to the next-hop switch, before sending it the call setup message. If a DS0 channel is available, the call processing engine provisions the circuit switch to forward data bits arriving at an assigned incoming timeslot on the input link to an assigned outgoing timeslot on the link to the next-hop switch. Call setup proceeds on a hop-by-hop basis to establish an end-to-end circuit. If a DS0 channel is unavailable on any hop, the call is blocked (rejected). If instead the circuit is set up successfully, phone conversation can proceed. When one of the users hangs up the phone, call release messages are exchanged between the switches on the end-to-end path to free up the DS0 channels. Each link in the POTS circuit network is modeled as an M/M/m/m queueing system in which calls are blocked when resources are unavailable [1]. There is no buffer available for holding calls. Since DS0 channels are required on all links for an end-to-end circuit, a call queueing approach could lead to poor utilization as assigned DS0 channels would be idle while waiting for DS0 channels on other links to become available. Therefore, calls are simply rejected if bandwidth is not available on any single link of the end-to-end path. Thus, POTS is a bufferless queueing system. In contrast to the POTS system, a relatively new service based on advance reservations (AR), which we refer to as Scheduled Dynamic Circuit Service (SDCS) [2] is being developed for Wavelength Division Multiplexed (WDM) and virtual-circuit networks (such as MultiProtocol Label Switched and Carrier Ethernet networks) [3]–[12]. SDCS has been deployed in backbone research-and-education networks such as ESnet [13], Internet2 [14], and commercial networks [15]. Requested circuit rates are typically high, e.g., 1 Gbps in a network with 10 Gbps links, for applications such as large dataset transfers. For example, ESnet interconnects scientific research laboratories whose users move large datasets from the supercomputing sites where they execute their parallelized scientific simulations to their home-institution storage clusters. SDCS is used to ensure guaranteed-rate service since IP- routed paths offer less predictable service. Circuit schedulers are required to support SDCS. Users and applications request circuits for immediate or future usage specifying the required circuit rate and duration (holding time), and circuit schedulers respond with an allocation or a rejection. Thus, SDCS uses a reservation system. To understand the conditions under which SDCS is more appropriate than a POTS-like service, we consider other exam- ples of reservation vs. queueing systems. For example, a doc- tor’s office is typically operated with a reservation scheduler, while grocery-store checkouts and bank tellers are operated as queueing systems. Also, airlines use a reservation scheduler, while city bus/train systems are operated as queueing systems. While for routine checkups, patients specify particular (future) dates/times for doctor appointments when making a reservation, often patients would like an earliest possible appointment for non-emergency care. Similarly, while airline customers typically specify dates/times when making a flight reservation, there are instances when they seek the earliest available flight to a destination.