Bin Packing with Linear Usage Costs – An Application to Energy Management in Data Centres

Abstract

EnergeTIC is a recent industrial research project carried out in Grenoble on optimizing energy consumption in data-centres. The efficient management of a data-centre involves minimizing energy costs while ensuring service quality. We study the problem formulation proposed by EnergeTIC. First, we focus on a key sub-problem: a bin packing problem with linear costs associated with the use of bins. We study lower bounds based on Linear Programming and extend the bin packing global constraint with cost information. Second, we present a column generation model for computing the lower bound on the original energy management problem where the pricing problem is essentially a cost-aware bin packing with side constraints. Third, we show that the industrial benchmark provided so far can be solved to near optimality using a large neighborhood search. The authors acknowledge their industrial partners (Bull, Schneider Electric, Business & Decision and UXP) as well as several public research institutions (G2Elab, G-SCOP and LIG). The authors from UCC are supported by Science Foundation Ireland Grant No. 10/IN.1/I3032. 3 Minalogic EnergeTIC is a Global competitive cluster located in Grenoble France and fostering research-led innovation in intelligent miniaturized products and solutions for industry.

Constraint Programming Based Large Neighbourhood Search for Energy Minimisation in Data Centres

“…In order to compare different upper bounds, we also computed lower bounds (LB) based on column generation with a 2 hour time-limit. The details of the lower bound computation is presented in [4].…”

“…We present a constraint programming-based large neighbourhood search (CP-LNS) for solving this problem which scales significantly beyond commercial optimisation tools such as CPLEX. 4 The key idea behind CP-LNS is to repeatedly consider a sub-problem of the overall problem and re-optimise it using constraint programming. We present a systematic empirical evaluation of our CP-LNS approach.…”

Delay and Reliability-Constrained VNF Placement on Mobile and Volatile 5G Infrastructure

“…Formulation 2 Bin Packing with Usage Cost [10] Input: VNFs V (G S ) as items with weight, host nodes V H (G I ) as bins with capacity Output: VNF placement respecting only capacity constraints…”

“…Next, the algorithm iteratively resolves the capacity, delay, battery and coverage constraint violations by changing the mapping location of VNFs in the initial invalid integer solution until a feasible mapping is found. First, we introduce the bin packing problem variation with variable bin and item sizes supporting linear usage costs [10] in Formulation 2. Lemma 1 (taken from [10] and pasted down for readability) states how to construct a fractional optimal solution for this bin packing variant, relaxing the integrality constraint.…”

Green Energy Aware Scheduling Problem in Virtualized Datacenters

“…Besides, in [3], Barrosso shows that over the lifetime of a data center, the expenses related to the energy consumption can easily surpass the hardware cost. In the literature, most work that attempts to reduce the energetic cost of a data center proceeds by reducing the overall energy consumption [9], [8]. Apart from the economic aspect, massive energy consumption also has repercussion on the environment, as the brown energy is produced from polluting sources.…”