2022 Optimization Days

HEC Montréal, Québec, Canada, 16 — 18 May 2022

Schedule Authors My Schedule

WC9 - Dynamic Optimization

May 18, 2022 03:30 PM – 05:10 PM

Location: Groupe Cholette (yellow)

Chaired by Bernard Fortz

3 Presentations

  • 03:30 PM - 03:55 PM

    An Integrated Cost and Availability Based Maintenance Scheduling Model for Data Centers Using Dynamic Programming Approach

    • Mostafa Fadaeefath Abadi, presenter, Concordia University
    • Fuzhan Nasiri, GERAD, HEC Montréal
    • Fariborz Haghighat, Concordia University

    Data Centers (DCs) have a large number of servers, computers and other equipment providing online and internet services for various companies in the Information Technology (IT) industry. In the last two years and because of the global COVID-19 crisis, the importance of DCs has been highly increased globally because of facing new realities and the high demand for online services in the world. Due to having several critical components, various requirements such as minimum availability, reliability, quality, and performance levels should be satisfied to assure the DC's business continuity and normal operation. In terms of maintenance management, the costs should be optimized to avoid over-maintenance and also, to meet the minimum system availability standards. Therefore, this abstract presents an availability-based optimization model, based on dynamic programming, to develop optimal maintenance scheduling for DCs to reduce their maintenance costs and fulfil the minimum availability requirements. This optimized maintenance management model could be advantageous for practitioners and can assist facility managers to manage the DC infrastructure more efficiently.

  • 03:55 PM - 04:20 PM

    Impact of equilibrium concepts on dynamic environmental agreements

    • Samar Garrab, presenter, Collège Militaire Royal du Canada

    In this paper, we study the impact of the signatories' leadership in a dynamic model, where countries take into consideration the impact of their emission decisions on the evolution of the pollution stock over time. We hence compare the stable coalition sizes as well as the equilibrium outcomes under two equilibrium concepts. Under the first one, all countries make their emission decisions simultaneously à la Nash. Under the second equilibrium concept, signatories act as game leaders and announce their emission strategies to defectors, who act as Stackelberg followers and decide about their emissions in reaction to this announcement. We solve a dynamic, discrete-time, infinite-horizon model where signatories can punish defectors at a cost. Our main findings are that the stable coalition size is lower under the Nash scenario compared to the Leader-follower scenario in the absence of punishment. Adding a punishment cost increases the stable coalition sizes under both scenarios, while the opposite occurs when the punishing cost increases. The impact of the pollution-stock farsightedness is positive from an environmental and economic perspective, leading to a lower pollution-stock steady-state, lower individual emissions for all countries, and higher long-term welfare.

  • 04:20 PM - 04:45 PM

    Online segment routing optimization considering polyhedral demand uncertainty

    • Bernard Fortz, presenter, Université Libre de Bruxelles
    • Jérôme De Boeck, Université Libre de Bruxelles
    • Stefan Schmid, Technical University Berlin

    The Internet is composed of a set of interconnected routers in which requests are routed along paths defined by routing protocols. Most intra-domain protocols which route requests along paths defined by routing protocols rely on shortest path routing (SPR): traffic flowing from a given source to a given destination will always be routed along the shortest path. The shortest paths are computed using a link metric system, in which weights are assigned to links.

    Even when metrics are optimized, SPR protocols suffer a number of draw-backs. In particular, restricting flows to shortest paths limits traffic engineering flexibilities. Furthermore, while a set of shortest paths may be efficient under certain demands, they may be suboptimal and lead to high congestion under other demands.

    To provide more flexibility Segment Routing (SR) was recently introduced. In a segment routed network, an ingress node may prepend a header to requests that contain a list of segments with routing instructions that are executed on subsequent nodes in the network. The headers containing routing informations can be dynamically updated in case a variation in demand occurs.

    We present stochastic and robust frameworks for online SR optimization considering a demand polyhedron. This polyhedron is defined by demands observed on the network. Our approach combines oblivious routing benefits with the flexibilities of SR. Numerical results compare our online procedure with existing oblivious SR.